The morphological productivity of selected ... - Helda - Helsinki.fi

The morphological productivity of selected combining forms in
English
Pro gradu thesis
Department of modern languages
University of Helsinki
26 September 2013
Eeva Rita-Kasari

Contents
1. Introduction ................................................................................................................ 1
2. Background ................................................................................................................ 4
2.1. Word-formation ....................................................................................................... 4
2.1.1. Basic concepts of word-formation ....................................................................... 4
2.1.2. The basic notions of the English word-formation system .................................... 6
2.1.2. The diachronic perspective on the English word-formation ................................ 8
2.2. Combining forms .................................................................................................. 10
2.2.1 Defining combining forms .................................................................................. 10
2.2.2. Neoclassical compounds .................................................................................... 13
2.2.3. The status of combining forms ........................................................................... 15
2.3. Morphological productivity .................................................................................. 18
2.3.1. Defining morphological productivity ................................................................. 18
2.3.2. Qualitative versus quantitative approaches to productivity ............................... 22
2.3.3. Measuring morphological productivity .............................................................. 23
2.3.3. Lexical statistics and productivity ...................................................................... 30
2.3.4. The psycholinguistic aspect of productivity ...................................................... 34
2.3.4. Constraints on productivity ................................................................................ 35
3. Material and methods ............................................................................................... 39
3.1. A closer look at the selected combining forms ..................................................... 39
3.2. Corpora versus dictionaries in the study of productivity ...................................... 42
3.3. How to determine units of analysis and other methodological problems ............. 45
4. Results ...................................................................................................................... 49
4.1. Potential productivity P ......................................................................................... 49
4.2. LNRE modelling ................................................................................................... 54
5. Discussion ................................................................................................................ 60
6. Conclusion ................................................................................................................ 65

Primary sources and software ...................................................................................... 68
References .................................................................................................................... 68

1. Introduction
Morphological productivity, the ability of speakers to coin new words using the
resources of a given language, has received a great deal of attention in the past
few decades, especially in the context of the English language. Bauer (2001) and
Plag (1999), for example, offer book-length accounts of this phenomenon.
According to Bauer, “productivity remains one of the most contested areas in the
study of word-formation” (1983: 62). Even though Bauer’s book was published
almost thirty years ago and many new theories on morphological productivity
have been proposed since then, this statement still seems to be valid. Even though
the concept of productivity seems to be easy to grasp intuitively, there are several
practical and theoretical issues that have not yet been tackled.
Baayen and his collaborators in particular have developed a number
of empirical corpus-based methods to gauge productivity quantitatively (see, e.g.,
Baayen 1992, 1993, 2009; Baayen and Lieber 1991; Chitashvili and Baayen 1993;
Baayen and Renouf 1996). In addition, the psycholinguistic aspect of
productivity, i.e., the way speakers store complex words in their mental lexicon,
has received considerable attention since the early nineties. Important
contributions have come from Frauenfelder and Schreuder (1992), Baayen (1993),
and Hay and Baayen (2002).
In addition to being an important theoretical concept, studying
morphological productivity also has practical relevance. Measuring degrees of
productivity is important from the point of view of writing grammars, for
example. While unproductive processes are listed in the lexicon, productive
patterns can be described by rules, and thus measuring productivity offers
valuable information on what processes should be given prominence in the case of
limited resources (Lüdeling et al. 2000: 57). The importance of the concept of
productivity can also be seen in the fact that computational tools cannot function
properly unless they take productive word-formation into account (Baayen 2008:
900). In addition, knowledge of the productivity of different word-formation
processes may also be relevant in language teaching. Since combining forms are
particularly frequent in scientific registers, this study might also have some
1

elevance in the context of English for special and academic purposes (see Fradin
2000: 22–23).
Studies concentrating on the productivity of English affixes abound.
Most of them have focused on suffixes (see, e.g., Baayen and Lieber 1991;
Baayen and Renouf 1996; Bauer 1992; Plag et al. 1999; Plag 2006, Palmer 2009,
Säily and Suomela 2009, Säily 2011). This might be due to several reasons, one of
which being their class-changing nature. Prefixation, on the other hand, has not
received the same amount of attention. Plag writes that in spite of the fact that
English is one of the best-described language in the world, “the exact properties of
many affixes are still not sufficiently well determined” (2003: 86), and that there
is still a great need for more detailed investigation.
Neoclassical compounding and what the Oxford English Dictionary
calls combining forms have been studied even less, and the few exceptions (e.g.,
Warren 1990, Prćić 2005, Prćić 2007, Prćić 2008, McCauley 2006, and Kastovsky
2009) have mainly been rather theoretical, and nothing really explicit has been
said about their productivity. The productivity of native N + N compounds has
been studied by Férnandez-Domínguez (2009), but neoclassical compounding
differs from them in many ways. This gap in research might partly be due to the
complex status of combining forms as a part of the English word-formation
system (see, e.g., Bauer 1998): they seem to be situated somewhere between
affixation and compounding (see section 2.2 for discussion).
To make the issue even more complicated, even the OED seems to
be rather inconsistent in the way it classifies certain elements as either combining
forms or traditional affixes. This inconsistency seems to concern initial combining
forms in particular. For example, the OED defines hyper- and -ultra as prefixes
and quasi- and pseudo- as combining forms, even though there seems to be no
apparent structural difference between these two groups of elements. It has also
been argued that the category of combining forms in general is superfluous and
that other, less marginal types of word-formation could well be used to describe
this word-formation process (Kastovsky 2009).
In this paper, I will measure the morphological productivity of
selected English initial combining forms (ICFs) and final combining forms
(FCFs), three of each, and see whether there are any differences between these
two rather closely related categories. The focus is on the so-called classical
2

combining forms, as opposed to modern combining forms (see Prćić 2005, 2007,
2008). The initial combining forms studied are hyper-, quasi-, and pseudo-, and
the final combining forms are -logy, -graphy, and -nomy. The study applies corpus
methods, and the data comes from the written part of the British National Corpus
(henceforth the BNC). My hypothesis is that the initial combining forms will
prove more productive and that they will combine more easily with native bases,
one of the reasons being that prefixes in general are often more lexeme-like than
suffixes, the latter often being used to convey grammatical information.
The main goal of this study is to clarify the status of combining
forms in the English word-formation system and to provide information on their
productivity. As for methodology, I will mainly rely on previous approaches that
have been widely tested in several studies. I will apply several indicators to assess
the morphological productivity of the selected combining forms. The first one is
the measure called productivity in the strict sense P (also called potential
productivity), developed by Baayen and his collaborators (see Baayen and Lieber
1991 in particular). In addition, two measures from lexical statistics, vocabulary
growth curves and frequency spectra, are employed. An open source statistics
software package, R, is used to obtain information on the selected combining
forms.
Section two provides the reader with the theoretical background
relevant for this study, defining and discussing the basic concepts related to
English word-formation. The category of combining forms is introduced, as well
as that of neoclassical compounds. The status of combining forms as a part of the
English word-formation system is of great relevance here, and it is therefore
thoroughly discussed. A brief history of diachronic approaches to the study of
affixation is also introduced. The notion of morphological productivity and the
theories attempting to describe it are also discussed, as well as the various
constraints that affect the application of various word-formation processes. In
addition, the basic theoretical background for lexical statistics and LNRE
modelling are introduced. Psycholinguistic approaches to productivity are also
discussed, since they have proved to capture an important aspect of the nature of
productivity. Furthermore, many quantitative measures of productivity are also
psycholinguistically motivated.
3

Section three discusses the method and the data chosen for the
present study, as well as potential problems that might turn up in the research
process. Important questions include the reliability of corpus evidence, as well as
the principles on which the units of analysis are determined. Section four presents
the results, which are then analysed and discussed in the fifth section. Section six
provides a brief summary of the most important findings of the thesis and
suggests some ideas for possible future studies.
2. Background
2.1. Word-formation
2.1.1. Basic concepts of word-formation
The concept of word is, in spite of its apparent simplicity, rather problematic in
linguistics. The notions of lexeme and word-form have proved much more useful
in the study of morphology. Haspelmath defines lexemes, or dictionary words, as
abstract entities that consist of several word-forms. Word-forms, or text words, on
the other hand, are the concrete realizations of a lexeme that can be pronounced
and used in texts (2002: 13). Plag uses the terms inflection and derivation to
distinguish between word-forms and lexemes: according to him, word-forms are
created with inflectional suffixes, while new lexemes are produced by derivational
affixes (2003: 14; see also Kastovsky 2001: 218). Since lexemes consist of several
word-forms, it is necessary to have a specific citation form, under which the
lexical entry is listed in dictionaries (Haspelmath 2002:
14).
From the point of view of morphological productivity, another
central distinction in morphology, the one between possible (or potential) words
and actual words, is also relevant. Plag defines possible words as words whose
structure follows the rules of the language. Actual words, on the other hand, are in
use in the language, although it is not clear how “in use” should be defined
(whether it be in the vocabulary of an individual speaker, or in a dictionary) (Plag
1999: 7). Plag writes that the usefulness of the possible-actual dichotomy has been
4

widely criticized, but he considers it a useful notion nevertheless: according to
him, studying large numbers of actual words is necessary in order to find out
about the mechanisms that are related to the properties of possible words (1999:
8–9).
The notion of mental lexicon is also relevant in the possible/actual
dichotomy. Anshen and Aronoff define the speaker’s mental lexicon as “the list of
irregular items that the speaker/hearer carries around in his or her head” (1998:
238). Words that are stored in the speaker’s mental lexicon are existing words,
while potential words are words that meet all the criteria for being a genuine word
in a language but that are not listed in the mental lexicon of that person (though
they may exist for another speaker) (Anshen and Aronoff 1998: 238). According
to Haspelmath, most morphologists agree that all simple and at least some
complex words are listed in speakers’ mental lexicons, but it is difficult to state
which complex words are listed and which are not, because most word-formation
processes cannot be described as either unequivocally productive or unproductive
(2002: 41).
Since this study applies corpus linguistic methods, another
distinction concerning the nature of words is worth mentioning, namely that of
types and tokens. In a sense, types represent different word-forms and tokens
different orthographic units or strings of characters in a text (see, e.g., Carstairs-
McCarthy 2002: 5–6). In other words, types are distinct word-forms, and tokens
all the running words in a text (Pápai 2004: 157). Thus, for example pseudoscience
and pseudo-sciences are considered as separate types, even though they
are strictly speaking manifestations of a single lexeme. Baayen and Lieber define
tokens as all the instances of a certain type (1991: 803). Both type and token
frequencies are always determined in relation to a particular corpus (Bauer 2001:
47). A common notation to describe the number of tokens, i.e., corpus size, is N,
while the vocabulary size, i.e., the number of types, is often referred to as V
(Baroni 2008: 805). It must also be noted that the number of tokens in a text or in
a corpus is always necessarily higher than the number of types in the same corpus.
5

2.1.2. The basic notions of the English word-formation system
There are two main processes of word-formation in English, namely affixation by
derivation and compounding. There are other, more marginal types of wordformation
in English, such as blending, clipping, and conversion, but this section
will mainly concentrate on derivation and compounding, since they bear the
strongest resemblance to the process of coining new words with combining forms.
Affixes are bound morphemes that are attached to bases, 1 often
modifying the meaning or the part of speech of the base in some way. They are
further divided into prefixes and suffixes: prefixes precede the base (co- in cooccur),
while suffixes are attached at the end of the base (-less in childless).
Minkova and Stockwell (2009: 101–102) state that affixes are relatively typical in
loanwords, since almost all the Greek and Latin loanwords in English contain one
or more affix. The third form of affix is called an infix, which is inserted into
another morpheme, as in abso-bloody-lutely (Plag 2003: 11). Infixes are generally
considered as a marginal type of word-formation in English (Plag 2003: 101).
With respect to semantics, the affixal meaning is often not as clear
as the meaning of the root (Minkova and Stockwell 2009: 71). Biber et al. state
that suffixes in particular often convey meanings that would be difficult to
paraphrase, a fact that might also be partially interconnected with their classchanging
nature. Prefixes, on the other hand, often express meanings that could
also be conveyed by using such things as adjectival premodifiers or other lexical
words (Biber et al. 1999: 324). For example, the meaning of such suffixes as -ist
or -ian can be explained by saying that they are used to form agentive nouns
(scientist, physician). On the other hand, the meaning of prefixes (e.g., neo-,
micro-) can often be explained by using adjectives (‘new’, ‘small’).
There are several ways to classify affixes in English, each of which
has its own advantages and disadvantages. Various classifications are introduced
in Plag (2003: 85–86). In addition to the obvious prefix-suffix dichotomy, he
mentions such criteria as the grammatical category of the base word, the semantic
properties of the affix, and the ability of the affix to change the grammatical
1 Root and stem are concepts that are easy to confuse, and in fact there is some disagreement as to
how they interrelate. Minkova and Stockwell define root as the central element of derivational
process that can either stand free (host) or be bound (seg in segment) (2009: 69). Base, on the other
hand, may either contain a root or a root plus affix, while the term stem is usually used in
inflection (2009: 71).
6

category of the base word. Plag states that one of the most unproblematic
classifications, at least with respect to suffixes, is the syntactic category of the
derived word. Some suffixes, such as -age and -ess, always result in nouns, while
-ify and -ize are used to derive verbs from nouns or adjectives. Prefixes, on the
other hand, are, because of their “more semantic” nature, often classified
according to their meaning (see, e.g., Plag 2003: 98–99, Marchand 1969: 134,
Minkova and Stockwell 2009: 102–105). As a final remark, it must be noted that
inflection, which allows the expression of various grammatical categories such as
number (with nouns) or tense (with verbs) also makes use of affixes, but it is
generally considered as part of syntax rather than word-formation (Plag 2003: 14).
Compounding, as opposed to derivation by affixation, involves
combining two (or possibly more) free forms into a single unit. In general,
compounding is the most important source of new lexemes in the English lexicon
(Minkova and Stockwell 2009: 9). As for their internal structure, compound
lexemes are usually divided into endocentric and exocentric compounds. 2
Endocentric compounds involve what is usually called a modifier-head structure,
where one element, the head, is somehow modified by another element (Plag
2003: 132). In English, compounds tend to be right-headed (Carstairs-McCarthy
2002: 61). For example, in doll house, house is the head that is modified by doll.
The head determines most of the semantic and syntactic properties of the
compound as a whole, so if the head is a noun (as in doll house), the whole
compound must be a noun (Plag 2003: 132).
Exocentric compounds, or bahuvrihi compounds, on the other hand,
exhibit a structure where the semantic head is outside the compound (Plag 2003:
145). For example, pickpocket is not a kind of pocket but a person who picks
pockets, and turncoat is not a kind of coat but a person betrays their original cause
by shifting to the opposing side. However, as Plag points out, these compounds do
have a head on the structural level, since the part of speech of the compound is
again determined by the right-hand member of the compound as a whole (2003:
145–146). Thus, pickpocket is a noun, not a verb. A third type of compounds
2 Carstairs-McCarthy uses the terms headed and headless compounds, respectively (2002: 64).
Minkova and Stockwell, on the other hand, distinguish between syntactic compounds, which are
formed by regular rules of grammar and are therefore not listed in dictionaries (e.g., birthplace),
and lexical compounds, the meaning of which cannot be computed from the meanings of its
constituent elements (ice cream) (2009: 10).
7

includes words such as singer-songwriter and poet-translator. They are called
copulative compounds, or dvandva compounds, and can be said to have two
semantic heads, neither of them being more prominent than the other (Plag 2003:
146).
2.1.2. The diachronic perspective on the English word-formation
The primary subject of this study is not the diachronic development of the English
word-formation system. However, a diachronic approach may explain some of the
differences between different word-formation processes in terms of morphological
productivity (see Carstairs-McCarthy 2002: 100). As Nevalainen has put it, the
rate of language change is most easily seen in the development of lexicon, despite
the fact that the core vocabulary of a language is usually relatively stable (1999:
332). None of the combining forms studied here are of native origin, but belong to
the foreign strata of the English lexicon.
Affixes have originally been independent words that have had an
auxiliary function, often derivational or inflectional (Warren 1990: 123–124).
Eventually, they have gone through a grammaticalization process, which has
resulted in semantic and phonetic erosion of varying degrees. Minkova and
Stockwell call this phenomenon semantic bleaching (2009: 102). For example, the
original meaning of the suffix -hood (‘condition’, ‘state of affairs’) is almost
completely bleached out and is almost impossible to specify in contemporary
English. Other similar examples in modern English include the native suffixesdom
and -ship, which are derived from nouns (Minkova and Stockwell 2009: 72).
On the other hand, sometimes the meaning of a word develops in the opposite
direction and a bound morpheme may acquire a new meaning as a free root, like
the noun ex (‘a former spouse’) (Minkova and Stockwell 2009: 72).
In general, the assumption is that new non-native affixes are born
when a sufficient number of complex lexemes are borrowed into a language and
they thus start to be interpreted as morphologically transparent (Nevalainen 1999:
377, Marchand 1969: 129). Transparency, which means the fact that a lexeme is
8

formally and semantically analysable into its constituent parts, seems to be at least
indirectly linked to productivity (Dalton-Puffer 1994: 252–253, Pounder 2000:
134). The relationship between transparency and productivity is discussed in more
detail in section 2.3.1.
In the course of history, the English vocabulary has been shaped by
several waves of loanwords from various languages, such as French, Latin, Greek,
and the so-called Neo-Latin, which was a mixture of Latin and Greek vocabulary
and which served as the lingua franca during the Middle Ages and the
Renaissance (Kastovsky 2009: 1). Besides the vocabulary, this considerable
borrowing has also affected the patterns of derivational morphology available to
speakers (Cowie and Dalton-Puffer 2002: 410).
In terms of such morphological strata, the lexicon of Old English
was very homogeneous, with only about three percent of the lexical items being of
non-native origin (Kastovsky 2006: 167). The Norman Conquest, however,
brought about enormous changes in the English word-formation system, and
many Germanic word-formation processes became obsolete (Marchand 1969:
130). These non-native derivational patterns became productive gradually
(Kastovsky 2006: 167–168). First, individual lexical items were borrowed, and
after a certain point “a formal-semantic relationship” could be established so that
the pattern could be extended to new formations by analogy, until the wordformation
process finally became productive.
The number of native prefixes in contemporary English is very small
(Marchand 1969: 129). They include a-, be-, fore-, mid-, mis-, and un-, and they
have their origin in independent words. During the Middle English period, a great
number of affixes, such as dis-, en-, inter-, non-, -age, -ation, -ive and -ous, was
borrowed from French and Latin, and by the Early Modern English period they
had become very frequent (Kastovsky 2006: 168). This abundance of borrowing
affected the whole word-formation system by causing competition, not only
between native and non-native affixes, but also between the non-native elements
themselves (e.g. the various negative prefixes a-, dis-, in- and non-) (Kastovsky
2006: 169). As for suffixes, the effect of the invasion of French, Latin and Greek
loanwords has not been as radical (Marchand 1969: 227).
During the Modern English period, the extensive borrowing has
continued, this time because of the development of scientific terminology,
9

introducing elements such as bi-, di-, hyper-, hypo-, meta-, micro- and multi- – the
number of prefixes outnumbering that of suffixes (Kastovsky 2006: 170). In
addition, any learned affixes, or combining forms, 3 have acquired new meanings
(Hughes 2004: 347). For example, extra- used to mean ‘out of’, as in
extraordinary (‘outside the scope of ordinary’) but is nowadays used to denote a
meaning ‘more than the usual’, as in extra strong. Similarly, super- ‘above’ and
ultra- ‘beyond’ are increasingly used to stress
size.
Kastovsky summarizes that the overall structure of English
vocabulary has changed and the number of derivational patterns available has
increased enormously from Old English to the present day, “partly rivalling each
other, partly restricting each other according to etymological domains” (2006:
170). In a similar vein, Hughes writes that different word-formation patterns are
used with greater flexibility in contemporary English than they were before:
recent coinages that exploit the suffix -able include, for example, such words as
doable, puttable and liveable (2000: 346).
2.2. Combining forms
2.2.1 Defining combining forms
It was stated above that affixes are always bound morphemes that attach to a root
or a base. What should then be thought about such words as neuro-logy or biology?
At first sight it seems that lexemes like these are formed with a prefix and a
suffix but no root, which of course runs counter to the basic assumption that
words should always contain a root (see Bauer 1983: 213–214).
Carstairs-McCarthy offers a partial solution by stating that affixes
are always bound but roots are not always free (2002: 20). Thus the first element
in words like audi-ence, magn-ify, or applic-ant is actually a bound root, since
they cannot occur alone in English (Castairs-McCarthy 2002: 19). The common
denominator with all these words is that they are all loanwords, borrowed either
3 The distinction between affixes and combining forms will be discussed in more detail in section
2.2.
10

from Latin or Latin via French. 4 The same “boundedness” applies to such
formations as neurology and biology as well, but in this case not only the first
element but also the final constituent is bound. This is what Castairs-McCarthy
means when he points out that words can in fact consist of more than one bound
root (2002: 21). In this sense, the constituents in words like neurology and biology
could also be called bound roots, but it has become a common practice by
linguists and dictionary-makers to call them neoclassical elements, or, more
commonly, combining forms (Plag 2003: 74; Carstairs-McCarthy 2002: 21).
The term combining form was first used in the New English
Dictionary (1884–1928), which was the predecessor of the Oxford English
Dictionary (Kastovsky 2009: 2; see McCauley 2006 for the use of the term
combining form in different editions of NED and OED). The current version of
the OED Online has a total of 2244 entries that are categorized as combining
forms. However, many entries under the label of combining forms have been first
published as early as in 1899, but have not yet been fully updated. As for the
definition of the term itself, the OED Online cites Bloch and Trager (1942: 66):
In Latin and other languages, many words have a special
combining form which appears only in compounds (or
only in compounds and derivatives)... The foreign-learned
part of the English vocabulary also shows a number of
special combining forms; cf. electro-, combining form of
electric, in such compounds as electromagnet.
For the third edition of the OED, the category of
combining forms is defined in the following way:
[A]n element used, either initially or finally, in
combination with another element to form a word. For the
purposes of OED3, a combining form differs from a prefix
or suffix by being generally noun-like or adjective-like
4 It is not always easy to differentiate between loanwords of Latin and French origin, partly
because these two languages are etymologically related and because it is not always clear at which
point in history a certain word has been borrowed into English. When borrowing from
Latin/French was at its most active phase, England was in fact a trilingual community, which
makes it even trickier to trace the etymology of some words (see McConchie 2006). Nevalainen in
fact uses the term latinate to describe words that are of either Latin or French origin (1999: 371).
11

and having a relatively full lexical meaning. (McCauley
2006: 96)
Marchand writes that the practice of coining new lexemes with
learned prefixes and combining forms dates back to the 16 th century (1969: 131).
This was mainly due to the Renaissance and the revival of sciences in Western
Europe, which produced an increasing need for scientific vocabulary (see also
Carstairs-McCarthy 2002: 66). However, the practice of coining new words with
combining forms did not become common until the 18 th and 19 th centuries, along
with the industrial revolution and the development of modern science that it
brought about (Marchand 1969: 131; Carstairs-McCarthy 2002: 66–67).
In general, the proportion of specialist terms in the lexical intake
grew steadily throughout the eighteenth century, which contributed to the fact that
neoclassical formations started to be associated with technical registers
(Nevalainen 1999: 365). The development of learned vocabulary usually goes
hand in hand with the development of the corresponding domains of science and
scholarship (Fradin 2000: 23). Quirk et al. point out that in Present-Day English,
in addition to scientific registers, combining forms are also common in
commercial brand names aimed at international markets (1985: 1535). Many
neoclassical compounds are also “in international currency”, which means that
they have been adopted in some form in several languages (Quirk et al. 1985:
1575).
It must be noted that there is some variation in terminology. So far
we have only talked about what has been usually called classical combining forms
(see, e.g., Fradin 2000: 11; Prćić 2005: 317). However, the term combining form
has also been extended to cover elements such as -gate in Watergate or -ware in
shareware (Lehrer 1990: 14, Fradin 2000: 11). Lehrer calls them “productive
splinters”, claiming that they have more in common with combining forms than
with affixes or roots (1990: 14). Prćić distinguishes between classical combining
forms, which have been adopted from Greek or Latin, and modern combining
forms, which are based on modern English words but which have been modified
to look like combining forms by, for example, the addition of a thematic vowel or
by clipping (2005: 317; 2007: 383). Examples include jazzo-phile) (from jazz), or
Thatcher-nomics (from economics). Another type of formation that Prćić defines
12

as combining forms includes various other bound forms that are independent
words in their source languages (e.g., scandal-monger and dulls-ville) (2007:
384). Prćić calls combinations like these, where one element is modern and the
other neoclassical (either with modern ICF + neoclassical FCF or neoclassical ICF
+ modern base) “semi-neoclassical compositions” (2005: 321).
This study will concentrate on what has been called classical
combining forms, which are said to represent the “core of established combining
forms” and are usually listed under the label of combining forms in dictionaries as
well (Fradin 2000: 12). Warren defines them as allomorphic variants of Greek or
Latin model words (1990: 115). They can either be derived from lexemes proper
(-logy) or Greek or Latin prepositions (hyper-, quasi-). Due to the limited scope of
this study, I have therefore excluded formations that are clippings from preexisting
words (e.g., euro-) and elements that are not strictly speaking new
morphemes at all (e.g., -gate; see Warren 1990: 115). The etymology of the
selected combining forms is discussed in more detail in section 3.1.
Bauer distinguishes between initial combining forms (ICF) and final
combining forms (FCF), depending on whether they occur in initial or final
position (2003: 214). This dichotomy has become widely adopted in studies
concentrating on combining forms (see, e.g., Kastovsky 2009). Quirk et al. write
that “the first element [of a compound] may be in its special ‘combining form’ (as
in the noun trouserleg or the adjective socioeconomic)”, and thus exclude final
combining forms from this definition (1985: 1567). In a similar vein, some
editions of the OED reserve the term combining form for the first element of a
compound and use the term terminal element for the final element (Kastovsky
2009: 4). However, this distinction has generally been replaced by the one
between ICFs and FCFs, which will be adopted in this study as well.
2.2.2. Neoclassical compounds
Formations that make use of combining forms are called neoclassical compounds
(Plag 2003: 74). Bauer mentions several characteristics that neoclassical
compounds have in common with native compounds in English (1998: 405). For
example, the meaning relationship between the constituent parts varies and cannot
13

always be “deduced in isolation from its context”. Secondly, they follow the
condition of hyponymy in the sense that they are hyponyms of their head element
(which is the rightmost element of the compound). These same word-formation
strategies were functional in the donor languages (Greek and occasionally Latin)
as well, the only difference thus being the fact that neoclassical compounds are
modern formations that are not attested in Greek and Latin as such (Bauer 1998:
405).
What makes neoclassical compounds atypical in English is that the
English language seems to prefer compounds that consist of free roots, such as
bookcase or motorbike (Carstairs-McCarthy 2002: 21). Combining forms, on the
other hand, have “little or no currency” as separate words, and they are not
normally the stressed part of a complex word (Quirk et al. 1985: 1520). Since they
consist of bound elements but otherwise behave like compounds, they have
sometimes been described as situated between compounding and affixation in the
English word-formation system (Quirk et al. 1985: 1520).
Many neoclassical compounds contain a central linking vowel,
usually -o-, 5 which cannot unequivocally be assigned to either of the constituent
combining forms (Carstairs-McCarthy 2002: 66). If the vowel is considered as
part of the first constituent, the final combining form in such words as
anthropology and cosmology will be -logy. However, in such words as toyology
(two occurrences in the BNC) or grandadology (one occurrence), the initial
element is of native origin, and thus the final element might be considered to be -
ology, which could then be considered as a morphologically conditioned
allomorph of -logy, in which case the central -o- will be part of the final
constituent (Kastovsky 2009: 6).
The primary stress of neoclassical compounds is usually on this
central -o-, a feature that distinguishes them from native compounds that usually
have the stress on the first element (Carstairs-McCarthy 2002: 66). In the majority
of cases, the -o- does not show up if the initial combining form ends with a vowel
or the final combining form begins with a vowel (e.g., tele-scope vs. laryngoscope)
(Plag 2003: 158). However, in some cases, such as in bio- and geo-, the -o-
5 Since the linking vowel -a- has been gaining ground especially in modern formations, Prćić calls
it a “modern linking vowel” (2007: 384). It must also be noted that the orthographically, other
forms are accepted as well, provided the appropriate phonetic criteria are met, as in bureaucracy.
14

seems to be part of the inherent structure of the combining form, hence bio-energy
and geoarchaeological (Plag 2003: 158). According to Kastovsky, the vowel has
originally been either a stem-formative or an inflectional ending (and thus
associated with the first element), but they have both become opaque in English
(2009: 6, 10). He suggests that the most useful interpretation of -o- would be as a
simple linking vowel.
Quirk et al. note that compounding with initial combining forms
may also be recursive, i.e., involve more than two elements, such as in
neurolymphomatosis or nephrolithotomy (1985: 1576). Final combining forms, on
the other hand, cannot be used recursively (Prćić 2007: 384). In terms of
semantics, the meaning of neoclassical compounds is usually predictable from the
meaning of its constituent combining forms (anthrop(o)- ‘human’ and , -(o)logy
‘science or study’ equal ‘science or study of human beings’), which is important
especially in coining new technical terms (Carstairs-McCarthy 2002: 66).
There are three ways in which a neoclassical compound may enter a
language (Cottez 1985: XVII, cited in Fradin 2000: 24). The lexeme may already
exist in the donor language, in which case it is borrowed with the same meaning
and corresponding form (phlebotomy vs. Greek φλεβοτομία). The second way is
adaptation: the lexeme exists in the donor language, but with a different meaning
(physiology vs. Greek φυσιολογία ‘natural sciences’). The third way is creating
new lexemes by combining elements that exist in the donor language as separate
units, not as a whole (chronology from χρονο and λογία).
2.2.3. The status of combining forms
Bauer argues that there is plenty of evidence for defining combining forms as a
category distinct from affixes: combining forms can combine with each other (an
ICF + FCF) but not with affxes, and they usually look more like lexical items than
affixes (1983: 214–215). The latter feature also applies to their phonological
structure, since they carry their own stress, contain full vowels, and are generally
disyllabic (Bauer 2003a: 35). In addition, affixes constitute a closed set of
morphemes, while combining forms form a relatively open set that grows when
15

new lexemes are created with the help of resources from Greek and Latin – or
English (Prćić 2005: 316).
Yet another feature that has been suggested as distinguishing
combining forms and affixes is that the same affix can never function as both
prefix and affix, while some combining forms can be seen both as ICF and FCF
(e.g., patho- in pathology -pathy in cardiopathy, both from Greek πάθος
‘emotion’) (Harastani et al. 2012: 74). On the other hand, combining forms also
have plenty of shared features with traditional affixes: for example, both can be
separated from the remaining element of the lexeme without their meaning or
form being changed, and both produce “output formations of a binary structure”
that are morphologically and semantically analysable into their parts (Prćić 2005:
315–316).
The line between combining forms and other non-native bound roots
is not always clear either: some combining forms can be combined with
traditional affixes as well (Carstairs-McCarthy 2002: 66). For example, socio- and
-logy can be used with the suffixes -al and -ic(al) to form such words as (psycho-
)social and logic(al). Prćić calls this phenomenon suffixal expansion, stating that
in such cases, the combining form becomes composite in nature. The added suffix
becomes the head of the combining form, and the word as a whole can be seen to
display a dual headedness where the head at the CF level (e.g., -ic) can be termed
the micro-head and the head at the word level (logic) a macro-head (2007: 386–
387). In this study, such suffixal expansions will not be taken into account in
analysing the selected final combining forms, for reasons explained in section 3.3.
To summarize, the status of combining forms in the English wordformation
system is not clear. They have features that distinguish them from
compounding on the native basis, but they are also distinct from affixes. The
problem is that neoclassical combining forms have not been studied very
systematically (see Kastovsky 2009: 1). Marchand criticizes the OED for being
inconsistent with the term and for not properly discussing their exact role in
English word-formation (1969: 132, 218). What he claims is certainly true, and
the OED’s, definitions for the category itself are vague at best.
However, Marchand himself is rather inconsistent in defining
combining forms and does not provide any solution as to their status. For
example, he writes that elements that are attached to full English words should be
16

termed prefixes, and others “are of a purely dictionary interest in any case” (1969:
132). This, as Kastovky has noted, is “a very questionable argument”, since they
are extremely frequent, especially in scientific registers (2009: 4).
Bauer argues that neoclassical compounding is in fact a rather
arbitrary category (they can be a part of clippings and blends, like in telethon, a
blend of television and marathon) and that it should be defined in terms of more
or less prototypical members of the category (1998: 407–409, 419; see also
Kastovsky 2009: 8). He suggests that problems with classifying neoclassical
compounds and combining forms might be due to the falsely unambiguous label
“neoclassical compound” itself (1998: 414). Kastovsky presents a similar view,
but goes even further by claiming that the term might as well be abandoned, since
“modern word-formation theory can well do without it” (2009: 2). He explains
this by the fact that even though modern English word-formation is dominated by
word-based morphology, 6 English has recently adopted some stem-based
morphology as well in the form of borrowed non-native elements, such as the
soci-al and logi-al mentioned above (2009: 9).
His solution thus involves simply introducing the concept of stem as
a possible lexeme representation (2009: 11), thus elaborating on the ideas
presented by Carstairs-McCarthy (2002: 21). The stems can either combine with
affixes (electr-ify) or with another stem (socio-logy), where the latter process
produces what has been called neoclassical compounds. As for the modern
combining forms, Kastovsky writes that such established word-formation
processes as blending (chunnel, from channel + tunnel) and clipping (euro-, as in
Eurocentric, eurocrat) can be used to account for them (2009: 11–12).
Kastovsky’s proposition is certainly appealing, and would make the
make-up of the English word-formation system much more simple and consistent.
It is, however, the ability of combining forms to combine with affixes that causes
some trouble. Bauer presents such ungrammatical examples as *electroness and
*electronization to justify the claim that the combinations are in general not
possible. It is true that such combinations electr-ic or soci-al are perfectly
grammatical, which would indicate the contrary. It must be also kept in mind that
6 In word-based morphology, the base form is directly accessed and thus not an abstraction, while
in stem-based morphology, the base does not occur without a derivational (or inflectional)
morpheme, such as in scient-ist (Kastovsky 2006: 157).
17

not all affixes can attach to all combining forms, just as it is not possible for any
affix to attach to any base. However, there are also some (usually initial)
combining forms that do not seem to be able to combine with any affixes, such as
pseudo- and hyper-. This might indicate that there is a cline between more affixlike
and more lexeme-like combining forms.
In addition, Kastovsky himself seems to be somewhat inconsistent
by terming such elements as auto-, hypo- and mono- as prefixes instead of
combining forms. But what is the exact quality that would make them prefixes,
not combining forms? The OED defines auto- and mono- as combining forms,
and hypo- as a prefix (the former being developed from lexical words, the latter
from a preposition).
If Kastovsky’s view is accepted, it will raise a number of questions.
What is the benefit of studying the productivity of a category that does not exist in
its own right? Still, the category of combining forms (at least in the classical
sense) might be useful from the point of view of the study of scientific
terminology. Furthermore, the productivity of these selected elements has not
been studied yet, and it may still reveal something new about them and help
situate them in the big picture of the English word-formation system. In addition,
it is the differences between initial and final combining forms that we are
interested in, and in this kind of study a term like combining form might not be
irrelevant. All in all, it can be said that none of the suggested solutions concerning
the status of combining forms presented in literature has proved completely
satisfactory so far. Thus, the term combining form will be used in this study,
partially for lack of a better term.
2.3. Morphological productivity
2.3.1. Defining morphological productivity
Morphological productivity is an important concept in the study of wordformation
and morphological theory in general (Cowie and Dalton-Puffer 2002:
411). Productivity is often defined as the speakers’ ability to create new lexemes
by exploiting the resources of a language. Speakers are able to create an almost
unlimited number of new words, and the lexicon of a language is thus never fixed
18

(Haspelmath 2002: 39). Therefore, word-formation processes can be called either
productive or unproductive, depending on the extent to which they can be applied
to form new morphologically complex words (Plag 2006: 537). 7
To illustrate the difference between productive and unproductive
word-formation processes, Plag mentions the verb-forming suffix -ize
(grammaticalize) and the prefix en- (enlist), which are both used to form verbs in
English (2006: 537). The former can be considered fairly productive, since it is
frequently used to form new lexemes that have not yet made their way in the
established lexicon. The latter, on the other hand, cannot be so easily used in new
formations in contemporary English, and it is thus felt to be unproductive. 8
Haspelmath writes that unproductive rules are a unique property of
morphology, since there is no direct equivalent to them in the area of syntax
(2002: 40). Carstairs-McCarthy explains this phenomenon by mentioning the
nonexistence of clearly ill-formed words (*ion-trans-al-at-form), but also of some
words that are in theory well-formed (*arrivation, *ridiculosity), as well as the
unpredictability and non-compositionality of the meaning of some existing words
(recital vs. recitation) (1992: 32). Bauer states that it is the productivity of wordformation
processes that is at least partially responsible for the huge vocabulary of
English, a fact that can be verified by consulting various dictionaries of
neologisms (1983: 63).
Another way to illustrate the difference between productive and
unproductive categories comes from Baayen, who defines productive categories
as ones with a growing membership, while unproductive categories have a fixed
or declining membership (2009: 900). Whether productivity should be understood
as an essentially synchronic or diachronic phenomenon (or both), is controversial,
however. Bauer claims that it is possible to talk about productivity in synchronic
terms, or about changes in productivity in diachronic terms, but not of
productivity as such as a diachronic phenomenon (1988: 61). Cowie and Dalton-
7 Terms such as word-formation process, word-formation rule (often abbreviated as WFR),
morphological rule, and morphological category are often used more or less synonymously in the
literature. They are usually used to refer to the way morphemes can be combined to form lexemes
(Plag 2003: 30, 179). The rule must contain information at least about the phonology and
semantics of the morpheme as well as about possible base morphemes (Plag 2003: 31). This is
what distinguishes “rules” or “processes” from the rather mechanical concept of morpheme. This
study will mainly use the term word-formation process.
8 One of the few exceptions is encrypt, which was first attested in the 1950’s, according to the
OED.
19

Puffer write that if productivity is to be taken as “a design feature of language in
general”, it should be understood as a synchronic notion (2002: 417).
On the other hand, it has been suggested that using type frequency,
for example, as an indicator of productivity (i.e., counting all the types formed
between two points in time) is inherently diachronic, even though the aim was to
study productivity from a synchronic point of view (see Cowie and Dalton-Puffer
2002: 418 for discussion). Be that as it may, changes in the productivity of certain
word-formation patterns have been frequently studied as well (see, e.g., Dalton-
Puffer 1994, 1996; Cowie 1998; Cowie and Dalton-Puffer 2002; and Säily and
Suomela 2009).
Plag asks whether productivity should be seen as a theoretical
primitive (i.e., an inherent property of word-formation rules) or whether it results
from other properties of these rules (2006: 537). According to Dressler and
Ladányi, productivity is “a primitive and prototypical property” of wordformation
rules, comparable to inflectional, syntactic, or phonological rules (2000:
119). On the other hand, Férnandez-Domínguez (2009: 53) claims that since
linguists often list a number of features associated with productivity (e.g., the
various constraints on it), the phenomenon might be considered as being of
composite nature and not an indivisible notion.
The distinction between inflection and derivation is also relevant in
defining morphological productivity. Although productivity is often considered
more relevant for the study of word-formation rather than that of inflectional
processes, Plag states that, at least to some extent, productivity also has to do with
inflection (2006: 538). For example, the regular past tense affix -ed might not be
considered fully productive, since some verbs form past tense forms with ablaut
or take no overt suffix, although the formation of the past tense is fully productive
as a category. According to Bauer, concatenative processes such as conversion
and back-formation should also be understood as morphological processes and
thus relevant for productivity as well (2005: 316). The productivity of compounds
has also been researched, though to a lesser extent (see Férnandez-Domínguez
2009). Bauer goes as far as to suggest that productivity even concerns syntax:
some syntactic structures are used more frequently than others to form new
sentences, and thus the notion of syntactic markedness might be understood as
some form of productivity or as relating to it (1995: 21).
20

Morphological productivity and linguistic creativity are two
concepts that are easy to confuse. There is in fact some confusion and overlapping
between the two terms among linguists themselves (see, e.g., Bauer 2001: 1, 62ff).
According to Bauer, productivity is an inherent feature of human language, which
allows speakers to exploit the rules of word-formation to create new lexemes,
while creativity allows for a native speaker “to extend the language system in a
motivated, but unpredictable (non-rule-governed) way” (1983: 63; see also Booij
2005: 67 and Dressler and Ladányi 2000: 105).
Renouf writes that creativity is often employed to achieve a stylistic
effect, and is typically seen as a departure from the rules of grammar, which can
manifest itself in such instances as punning or wilful error (2007: 70). Evert and
Lüdeling state that words that result from productive processes are often not
recognized as new by speakers, while creative formations are perceived as new
(2001: 475). Bauer also uses the term “semi-productivity” to describe linguistic
creativity, claiming that it differs from productivity by requiring some sort of
encyclopaedic, extra-linguistic knowledge (1995: 26, 28). Plag writes that
productivity is often considered unintentional and creativity intentional, but he
also claims that the intentional/unintentional dichotomy is not very useful, since
speakers have varying levels of awareness of the structure of the lexicon. In
addition, even productive rules may be used intentionally (1999: 13–14).
It must be noted, however, that productivity should not be confused
with regularity either, since all productive patterns are rule-governed and thus
regular, but so are many unproductive rules as well (Dressler and Ladányi 2000:
105). For example, the suffix -ess is perfectly regular, but cannot be used to coin
neologisms (Haspelmath 2002: 111). The same holds true for transparency: a
lexeme can be perfectly transparent but not necessarily productive (Pounder 2000:
134–135; Bauer 2001: 54). 9 On the other hand, productive processes are usually
formally and semantically transparent, unless the process becomes lexicalized
(Bauer 2001: 135). Dalton-Puffer points out that the term transparency is often
9 Transparency and regularity are closely related concepts, and sometimes used synonymously.
Pounder (2000: 135) suggests that transparent emphasizes the “analysis side of morphology”,
while regular has more to do with the fact that the formation is not isolated in the morphological
system of a language from the speaker’s point of view. Another related concept is analysability,
which is used by Dalton-Puffer 1994, who uses it more or less synonymously with transparency.
21

used in a vague sense in the literature, since it is usually not specified whether the
focus is on the linguist’s or the speaker’s point of view (1994: 253).
2.3.2. Qualitative versus quantitative approaches to productivity
The approaches to defining the nature of productivity can roughly be divided into
two, qualitative and quantitative. If productivity is considered a qualitative
phenomenon, a given word-formation process either has this character or not
(Plag 2006: 538–539). The qualitative approach to productivity has mainly
concentrated on various kinds of phonological, morphological, syntactic and
semantic restrictions of different word-formation processes, the goal having been
to provide an intensional description of all the possible bases of a word-formation
process (Evert and Lüdeling 2000: 167). The most typical constraints on the
productivity of word-formation processes are discussed in section 2.3.3.
It has also been claimed that the productivity of word-formation
processes can be described as a continuum rather than as an all-or-nothing
property in the sense that these processes are productive to a varying extent (Plag
2006: 539). Baayen calls this kind of productivity scalar, as opposed to an
absolute view of productivity (1992: 185). Most scholars currently seem to
assume at least some degree of scalar productivity, whether it means that
productivity is infinitely variable on a scale or that there is a certain number of
steps on that scale (see, e.g., Bauer 2001: 15–16, 126). Even though the
qualitative and quantitative approaches to productivity can be seen as a
dichotomy, the suggestion is that they are nevertheless closely related. As Plag
has put it, “the idea of potentiality, which is central to qualitative definitions of
productivity, can be expressed in the statistical terms of probability” (1999: 22).
The distinction between the availability and profitability of
morphological processes is quite useful in grasping the difference between
qualitative and quantitative aspects of productivity. The terms come originally
from Corbin, who calls them disponibilité and rentabilité, respectively (1987:
177). 10 The availability of a word-formation process means whether it can be used
10 The English translation is by Carstairs-McCarthy (1992: 37).
22

to form new lexemes, and the profitability of a process shows how many bases the
process can affect (Carstairs-McCarthy 1992: 37). Availability is thus a
qualitative notion, while profitability can be seen as quantitative (Plag 2006: 122).
Profitability is often used as synonym for type frequency, that is, the
number of words that have been produced using a certain word-formation process
(the number of actual words). In this sense, profitability is a rather straightforward
concept, since it is possible to test it empirically (Férnandez-Dominguez 2007:
60). Availability, on the other hand, is a trickier concept. Again, the problem lies
in the synchronic versus diachronic dichotomy. Processes that are being used, i.e.,
available, at present, are rather unproblematic, but problems arise with processes
that have ceased to be used productively. Férnandez-Domínguez suggests that
availability corresponds to the paradigmatic axis of productivity and profitability
to its syntagmatic axis (2007: 63–64). In other words, availability concerns the
selection between competing suffixes, such as the noun-forming suffixes -ance, -
al, or -ion.
2.3.3. Measuring morphological productivity
As was stated above, quantitative approaches to productivity often assume that
there is a way to operationalize the problem and determine the exact degree of
productivity of a particular word-formation process. According to this view, the
productivity values of different processes can also be compared. Productivity as a
quantitative phenomenon is probably mentioned first by Bolinger, who defines it
as “the statistically determinable readiness with which an element enters into new
combinations” (1948: 18). Since then a number of methods to measure
productivity have been proposed. The most important measures will be discussed
next.
Type frequency has been suggested by several authors as a rather
good measure for gauging productivity. Baayen calls this kind of measure realized
productivity (2009: 901–902) or extent of use (1992). Haspelmath uses the term
degree of generalization (2002: 109). The basic assumption with realized
productivity is that a productive morphological category has many members.
Thus, it can be measured simply by counting the number of different types (that
is, different word forms) within a certain morphological category.
23

The method is not devoid of problems, however. Type frequency
does not tell us anything about the availability of a process, i.e., whether the
process can be used to form new words in the future or whether it has been
“saturated” (Bauer 2001: 144–145). For example, the suffix -ment has a high type
frequency, but there are only a few neologisms with -ment in the OED that have
been recorded in the twentieth century (Haspelmath 2002: 109). In this sense, type
frequency can only be used as an indicator of past productivity (Cowie and
Dalton-Puffer 2002: 416). In addition, if type frequency is assumed to equal the
degree of productivity of a certain word-formation process, it may lead to
counterintuitive results. There are many classes of words that have several
members but that are not productive, such as pronouns.
Type frequency has been used as an indicator of productivity,
especially in historical studies (as in Dalton-Puffer 1996, Cowie and Dalton-
Puffer 2002), since many statistical methods do not function properly if there is
scarcely data available (see below for discussion). Another method used in
diachronic studies is counting the number of neologisms that are attested during a
specified period of time (Plag 2006: 541). The neologisms can be determined with
the help of a good historical dictionary, but the reliability of the dictionary may
become an issue. For example, lexicographers may overlook new words that are
formed with a very productive pattern (Haspelmath 2002: 110). Bauer also
mentions the danger of confusing the actual process of word-formation from the
results of the word-formation: in English, for example, there are several complex
words (such as payable) which are not created in English but are loans (2001:
145). The advantages and disadvantages of dictionaries, as opposed to corpora,
are discussed in section 3.2.
A more refined view of measuring productivity comes from Aronoff
(1976), who has proposed that productivity should be understood as a ratio of
actual to possible words: the higher the ratio, the greater the productivity of a
given word-formation rule. However, it is not a simple task to count the number
of possible words with that word-formation process. It could, in theory, be
determined by identifying all the possible restrictions on that pattern, but in
practice there are several unproductive rules in English that do not seem to have
any identifiable restrictions (Haspelmath 2002: 110). In addition, since the
possible bases may themselves be complex words that are formed productively,
24

the set of possible words becomes practically infinite (Haspelmath 2002: 110).
Anshen and Aronoff have also found out that for highly productive wordformation
processes, the model gives a rather low productivity index, since the
number of possible words is, in theory, infinite (1981: 64). The most fundamental
problem with Aronoff’s model, though, is how to actually count the number of
possible words. Baayen (1992, 1993) has attempted to formalize Aronoff’s model
by introducing the potentiality measure I, which is the ratio of the number of
possible words S and the number of actual words V (the reverse of Aronoff’s
index) with a given affix.
The methods discussed above can be defined as quantifications, as
opposed to probabilistic measures (see Plag 1999: 24). Baayen and his
collaborators in particular (see, e.g.,Baayen 1992, 1993, 2009; Baayen and Lieber
1991;Chitashvili and Baayen 1993; Baayen and Renouf 1996) have developed a
number of statistical methods to measure productivity that make use of large textcorpora.
Most of the statistical measures to gauge productivity rely on the
notion of the hapax legomenon. The words hapax legomenon constitute a Greek
expression that has been adopted from classical studies, and it literally means
‘said (only) once’ (Carstairs-McCarthy 2002: 96). Hapax legomena (or just
‘hapaxes’) are words that occur only once in a certain corpus (Bauer 2005: 325).
With productive morphological processes a large number of low
frequency words is to be expected, because our mental lexicon “allows the
decomposition of the newly encountered word ... and thus the computation of the
meaning on the basis of the meaning of the parts” (Plag 2006: 542). Chitashvili
and Baayen call this kind of phenomenon a Large Number of Rare Events
(LNRE) distribution (1993: 57). 11 Unproductive processes, on the other hand, tend
to result in a large number of high-frequency words (Plag 2006: 542). In a similar
vein, the fact that a productive word-formation rule is available in the mental
lexicon guarantees that all complex words with that affix can be produced or
understood, even if they have a very low frequency (Baayen and Renouf 1996:
74). This implies that hapaxes can be said to be a good indicator of productivity.
11 How this distribution can be further used in the statistical study of productivity will be discussed
in section 2.3.3.
25

It is not important whether the hapax legomena are familiar to the
speaker or not, but rather the fact that they “represent the rate at which new words
are being coined in the language as a whole” (Bauer 2005: 325). 12 It is also worth
noting that not all hapax legomena are necessarily neologisms – while some
hapaxes are just old or obsolete words, some neologisms may occur more than
once in a corpus or be completely unattested (Szymanek 2005: 430–431). Thus,
hapaxes are not a goal in itself: they merely function as a statistical tool for
estimating productivity (Baayen 2009: 905–906).
The most famous and widely cited of Baayen’s measures also makes
use of hapaxes. The measure has several names; most often it is called potential
productivity (Baayen 2009: 901–902), productivity in the strict sense (Baayen and
Lieber 1991: 821), or the category-conditioned degree of productivity (Baayen
1993: 200). Baayen and Lieber (1991: 809) have formalized the measurement of
potential productivity as follows:
where n 1 represents the number of hapax legomena with a given affix and N the
number of tokens of all words with the same affix. The resultant figure is a
decimal that has a value between 0 and 1. The higher the value, the more
productive the process is, and vice versa. For example, if there are 100 tokens
with the prefix re- in the corpus, and 6 types that occur only once, i.e., are
hapaxes, the P value of re- is 0.06. For the sake of simplicity, this measure will be
mostly referred to as P.
Baayen and Lieber state that P allows measuring “the rate at which
new types are to be expected to appear when N tokens have been sampled” and
estimating the probability of encountering new types (1991: 809). Therefore, if
the sample is representative enough, “P makes a statement of what potentially
12 Cowie and Dalton-Puffer state that the”newness” of words is usually defined in terms of
dictionaries:”what is not in a given comprehensive dictionary is taken to be new”. This must not be
confused with the psycholinguistic aspect of newness, since a word that is well established in a
dictionary may be new to individual speakers (2002: 419).
26

could have been in the sample but has not been actualized for some reason or
other” (1991: 811). In other words, the measure captures the probability that the
next token in the sample is a hapax: according to Baroni, the proportion of
hapaxes at the Nth token should be a good enough estimate of how likely a word
N + 1 will be a hapax (2008: 818).
Baayen and Renouf state that if the size of the corpus is increased,
the number of hapaxes increases as well, but not in a linear way (1996: 75). This
is explained by the fact that while the sample is still small, the probability of
encountering a so far unattested word is relatively high. On the other hand, the
larger the sample is, the more likely it already contains that particular word. Thus,
the measure does not allow the comparison of P figures between different corpora
(Baayen 1993: 191). 13 The growth rate of the vocabulary and hence the slope of
the vocabulary growth curve varies at different sample sizes. Vocabulary growth
curves will be discussed in section 2.3.3.
Potential productivity is not the only statistical method developed by
Baayen and his colleagues. The hapax-conditioned degree of productivity P* was
introduced in Baayen (1993). Also called expanding productivity, it captures the
rate at which a category is “expanding and attracting new members”, and is
calculated by dividing the number of hapaxes of a given category by the number
of all the hapaxes in a corpus (Baayen 2009: 901–902). In other words, P* can be
used to study the contribution of a certain word-formation process to the
expansion of the vocabulary in general. Baayen suggests that P and P* should be
seen as complementary processes, the former being used to distinguish between
productive (even though this is not simple at all) and unproductive processes and
the latter to rank productive processes (1993: 194).
The measures introduced in the work of Baayen and his colleagues
are not the only statistical methods that have been developed to gauge
productivity. Säily and Suomela (2009) have designed a model that makes use of
accumulation curves and statistical significance testing. In their method, the
corpus is divided into samples that preserve discourse structure. A growth curve
for a morphological category (such as a particular affix) is then plotted on a
13 This is a specific instance of a more general problem of “variable constants” in lexical statistics
(Tweedie and Baayen 1998), cf. type/token ratio.
27

figure. 14 A sample is picked randomly, the number of types or hapaxes is
calculated, and the sample is plotted on a figure (see Figure 1). Another (random)
sample is picked and added to the previous one, until the whole corpus has been
sampled. The procedure is then repeated, for example, a million times with the
help of a suitable computer software, and the permutation, i.e., the order of the
samples changes every time. The result looks something like Figure 2: it is
possible to indicate visually the area covered by a certain percentage of the
curves, which helps us to see the extent to which the productivity rates of a certain
feature in a sub-corpus differ from those of the corpus as a whole (Säily 2011:
125).
According to Säily, type accumulation curves are a particularly good
measure in comparing productivity between different (e.g., gender-specific)
subcorpora (2011: 127, 135). There are two ways to plot a type accumulation
curve: either the number of types as a function of corpus size, or the number of
hapaxes as a function of the number of types. Hapax accumulation curves,
however, require a large corpus in order to produce statistically valuable data in
which the results are not skewed (Säily 2011).
14 The curves can be plotted either for types (as a function of token frequencies) or hapaxes (as a
function of running words in the corpus) (see Säily 2011: 126).
28

Figure 1.Two type accumulation curves. Each tick mark represents the addition
of one sample. From Säily and Suomela (2009: 102).
Figure 2.Bounds for 1,000,000 accumulation curves for the suffix –ity in the 17 th -
century part of the CEEC.Gender-based subcorpora plotted on the curves, which
show that women have a significantly lower type frequency. From Säily and
Suomela (2009: 100, as cited in Säily 2011: 126).
29

2.3.3. Lexical statistics and productivity
As was shown with the example of type accumulation curves, the study of
productivity can benefit from what is usually called lexical statistics, i.e., the
study of word frequency distributions and the probabilities of occurrence of
different types across texts and corpora (see Baroni 2008: 803). How different
word types are distributed in texts usually follows some pattern. For example, it
was pointed out above that productive word-formation processes are characterized
by a large number of low-frequency words, such as hapax legomena or dis
legomena (words that occur twice in a corpus). Interestingly, samples consisting
of naturally occurring text, independently of size, language, or textual typology,
are characterized by a similar pattern of few very frequent words (typically
function words such as a, the, of, etc.) and a long tail of low-frequency types
(Baroni 2008: 810). A statistical model like this is called a “large number of rare
events” (LNRE) distribution (Baayen 2001: 51).
Such a skewed distribution with “few giants ... and an army of
dwarves” (Baroni 2008: 814) can be predicted by Zipf’s law (Zipf 1949, 1965). 15
The model shows how frequency is a non-linearily decreasing function of the rank
of a word in the text. For example, the most frequent word in the text has the rank
1, the second most frequent word has 2, and so on. Zipf’s law is formalized as
follows:
( )
( )
where f(w) is the frequency of a word w, and r(w) the rank of the same word,
while C and a are constants that depend on the corpus (Baroni 2008: 813). If a is
assumed to be 1, then C is the frequency of the most frequent word in a corpus
(having thus rank 1). If the most frequent word occurs 3000 times in the corpus,
then C = 3000. The second most frequent word then has a frequency of C/2 =
1500, the 100 th most frequent word C/100 = 3, and so on. Thus the frequencies of
words decrease rapidly among the high ranks, but slows down as the rank
decreases.
15 Due to the limited scope of the study, the deep mathematics underlying the Zipf-Mandelbrot
model are discussed here only cursorily. A more thorough account of the mathematical motivation
of the model is given by Baayen (2001).
30

Even though Zipf’s law is sound in theory, it does not provide a
perfect fit for expected values of word frequencies (Baroni 2008: 815). A better fit
is obtained when the extra parameter b is added to the formula:
( )
( ( ) )
The resulting formula is called Zipf-Mandelbrot’s law, of which the original
Zipf’s law is a special case with b = 0 (Baroni 2008: 815). The optimal parameters
can be determined with the help of a computer. The zipfR package, available for
the R software, has been used in this study (see Evert and Baroni 2007).
Table 1. A toy rank/frequency profile (left panel) and frequency spectrum (right
panel)
r f f V(f)
1 13 1 10
2 7 2 5
3 5 3 3
4 2 4 2
5 1 5 1
Baroni mentions two useful tools that help capture word frequency
distributions in a visual way: rank/frequency profiles and frequency spectra
(2008: 806). A rank/frequency profile resembles a type frequency list on which
the types in a corpus and their respective frequencies are placed in a descending
order. In a rank/frequency profile, the type labels are replaced by their frequencybased
ranks, 1 referring to the most frequent type and so on. A frequency
spectrum, on the other hand, reports how many types have a certain frequency.
Table 1 offers an example of a rank/frequency profile and a frequency spectrum
31

which can be seen as reverse sides of the same phenomenon. A more visual way
to illustrate those two frequency distributions is to present them as a bar plot
graph. An example is offered in figure 3, which draws from data for the present
study.
Figure 3. A rank/frequency profile and a frequency spectrum for the initial
combining form hyper- in the written part of the BNC.
What makes rank/frequency profiles and frequency spectra
interesting from the point of view of productivity is that if we know the
distribution of type probabilities of a certain word-formation process, the expected
vocabulary size V and the number of hapax legomena V 1 can be calculated for any
sample size N (Baayen 2001: 41–51). Since potential productivity P essentially
measures the vocabulary growth rate of different processes and consequently the
slope of the growth curves for that process, comparing the slopes will give us
information about the differences in productivity between different processes. The
vocabulary growth of a process can be visualized by vocabulary growth curves,
which are essentially the same thing as the type accumulation curves (Säily and
Suomela 2009, Säily 2011), displaying the type frequency V of a certain wordformation
process as the function of its token frequency N. An exemplary
vocabulary growth curve is presented in Figure 4.
32

Figure 4. Vocabulary growth curves for the Italian prefixes ri- (interpolated) and
ultra- (extrapolated). The x axis displays the corpus size N and the y axis the
number of types V in relation to N. From Evert and Baroni 2007.
However, since the vocabulary does not grow linearily, direct
comparison of differently sizes samples, i.e., different word-formation processes,
is not reliable. In the case of samples of different sizes, it is therefore necessary to
extrapolate the growth curve of the smaller sample and predict the growth rate for
an arbitrary sample size. This involves computing the expected sample size for a
smaller corpus in order to “stretch” it to the length of a larger corpus (Baroni
2008: 820). This can be done with the zipfR package for the R statistics software.
16
Figure 4 (from Evert and Baroni 2007) illustrates the difference
between the productivity of two Italian prefixes. The prefix ultra- has a smaller
population that the prefix ri-, but when the sample is extrapolated to equal the size
16 For a more detailed account of the zipfR package and its implementation, see Evert and Baroni
2007.
33

of ri-, and the resulting vocabulary growth curve is compared to that of ri-, it can
be seen that it in fact has a steeper curve. This tells us that it has a higher growth
rate, i.e., a higher P value. In general, it is typical for a vocabulary growth curve
to grow more steeply at lower sample sizes, because if a corpus is small, it is more
likely to encounter new, unattested types. As the corpus grows, the likelihood
decreases.
2.3.4. The psycholinguistic aspect of productivity
Attempts to measure degrees of productivity constitute only one aspect in the
study of this phenomenon. Another issue is how speakers process, storage,
retrieve and produce morphologically complex words in terms of their mental
lexicon (cf. Bauer 2001: 101, 112), and what the role of productivity is in this
schema. The primary interest of this study is not the psycholinguistic aspect of
productivity, but it is worth discussing nevertheless, since many quantitative
measures of productivity are also psycholinguistically motivated, as will be shown
below.
A crucial question at this point is whether complex words are stored
as single units or as distinct morphemes in the mental lexicon, while the latter way
involving morphological parsing. 17 The former way is more adequate from the
point of view of processing, the latter facilitates the storage of the forms
(Frauenfelder and Schreuder 1992: 166–167). In Frauenfelder and Schreuder’s
model, each complex word is simultaneously processed in both ways, the faster
route winning the race (hence the term morphological race model) (1992: 175).
Factors that influence which route will win include token frequency (high token
frequency favours the direct route) and transparency (1992: 176–177). The model
is related to Baayen’s statistical productivity measures by predicting that the
parsing route wins with transparent, infrequent words – the very properties of
words that are assumed to be of high productivity in Baayen’s model.
Relying on Frauenfelder and Schreuder’s work, Baayen (1993)
discusses the activation level A, a measure that attempts to determine which route
17 Parsing has several different (although related) senses. In this context, by parsing one means the
way speakers analyse complex words into their constituent parts (see, e.g., Frauenfelder and
Schreuder 1992: 175).
34

to choose in the race. The activation level of a word-formation process is
formalized as the number of tokens containing that process with a frequency less
than a threshold θ (1993: 196). If the frequency is higher than the threshold, the
word is processed by the direct route, and while words with a lower frequency are
parsed (1993: 196). For example, in the present data, the activation level for
pseudo- would be 491, if θ = 8 (the same as Baayen uses). Thus, all types with a
frequency less than 8 are parsed. Baayen does not tell us how exactly the
threshold θ should be determined, but he writes that if the value of θ is high, more
opaque forms will be included in the frequency counts (1993: 203).
Baayen’s model is further improved by Hay (2001, 2003), who
states that instead of the absolute frequency of a word, the degree of
decomposability of complex words depends on the relative frequency of the
derived word and the base, i.e., the frequency of the derived word divided by the
frequency of its base. For example, the derivative business is more frequent than
its base busy, which means that it has a high relative frequency and that it is
accessed as whole (Plag 2006: 548). By contrast, words with a low relative
frequency are often morphologically productive and semantically transparent.
Plag summarizes that the semantical transparency of productive word-formation
rules is the consequence of cognitive processing (2006: 548). Hay and Baayen
(2002) attempt to further refine the model by defining the specific parsing line
above which derived words are most likely to be parsed.
Neither Baayen’s activation level nor Hay’s relative frequency
model will be applied in this study as such. However, psycholinguistic evidence
might prove useful in determining the appropriate units of analysis, i.e., how to
deal with pseudo-affixed words. Baayen actually offers some guidelines on how
to determine appropriate units of analysis when studying productivity (1993: 200–
201). Questions like this will be further discussed in section 4.
2.3.4. Constraints on productivity
A word-formation pattern cannot apply to any word in a language, but only to a
certain subset of words (Rainer 2005: 335). In other words, there are various
factors that restrict the potential combinations of affixes. It has even been
proposed that all morphological rules are equally productive, so that the
35

differences in (apparent) productivity are caused by restrictions on the rules.
However, there are many rules that do not seem to be restricted in any general
way, which corroborates this idea. 18 The study of the various restrictions on wordformation
processes is related to the qualitative aspect of productivity.
The subset of possible bases for a word-formation process is called a
domain (Rainer 2005: 335). This phenomenon can be called a restriction, but most
scholars use the term constraint instead. Bauer, for example, justifies the use of
constraint by the fact that the restrictions are not always absolute (2001: 126).
There are several ways to classify different kinds of constraints. Rainer, for
example, divides them into universal and language-specific (2005: 335). The
former apply to all languages, while the latter depend on the language in question.
Another division of constraints is often made between external
factors (factors that are mainly related to fashion, knowledge, and attitudes) and
internal factors (caused by the internal structure of a language) (Férnandez-
Domínguez 2007: 69). I will mainly follow Rainer’s model, although my focus
will be on the language-specific constraint relevant for the English language and
especially on the various structural constraints that comprise what Férnandez-
Domínguez calls “all the levels of linguistic description” (2007: 69). All the
structural constraints (i.e., phonological, morphological, syntactic, semantic,
pragmatic) are thus language-specific, and the only constraints treated here that
fall under the category of universal constraints are the various types of blocking.
Phonological constraints on morphological compatibility are a very
common type, illustrated by phonologically conditioned allomorphs, such as the
complementary set of plural markers /iz/, /s/ and /z/ (Bauer 2001: 128). Bauer
mentions three properties of the base that may constrain what affixes may attach
to it (2001: 129–130). Sometimes the segmental make-up of the base is important:
the verb-forming -en only attaches to bases that end in a stop or a fricative). In
other cases the suprasegmental make-up, i.e., stress, is what matters: the nounforming
suffix -al is only attached to bases that have the stress on the first
syllable. Sometimes the number of syllables in the base determines which affixes
may attach to it. As regards the third case, Bauer does not mention any examples
18 For discussion, see Haspelmath (2002: 111).
36

in English. Phonological restrictions are a common example of language-specific
constraints.
With morphological constraints on productivity, the morphological
structure of the base determines what kinds of affixes may attach to it (Bauer
2001: 130). Again, Bauer lists three options (2001: 130–132). Some affixes apply
only to a certain morphologically defined set of bases. The set can be defined in
terms of etymology, for example. This is often called level ordering (see below
for discussion). Sometimes a certain affix may be added only to a derived or to an
underived base. The base may also have to end in a particular affix before any
further affixes can be added. In general, prefixes have fewer word-class
restrictions than suffixes, because they do not usually alter the word-class of the
base they attach to (Nevalainen 1999: 355).
The distinction between morphological and syntactic constraints is
not always clear-cut. Bauer draws the line by defining morphological constraints
as ones that concern the internal structure of the word, while syntactic constraints
deal with the way a word is used in context (2001: 133). A typical example of a
syntactic constraint is the fact that many affixes can only be added to a certain
word class as a base (Bauer 2001: 133). Sometimes the transitivity of the verb
base is of importance: for example, the suffix -able may only be used on transitive
verbs.
Level ordering is one of the language-specific constraints on
productivity. Level ordering arises because many languages have several
morphological strata, a native and a non-native one, which both have some
combinatory restrictions (Rainer 2005: 340). For example, in English, many
Latinate affixes are restricted to Latinate bases only (Haspelmath 2002: 108).
However, since speakers do not necessarily know whether a stem belongs to the
native or non-native stratum, the restriction on the basis of morphological strata is
sometimes unstable: the suffix -ous has been added to some native bases as well
(murderous, thunderous) (Haspelmath 2002: 108).
Finally, pragmatic and sociolinguistic factors may also restrict the
productivity of word-formation processes. While semantic constraints depend on
the linguistic nature of the base, pragmatic constraints (as opposed to structural
constraints) involve the way in which words are actually used (Bauer 2001: 135).
Certain patterns are only used in particular situations, and it requires language
37

competence to know in which situations to use which patterns (Rainer 2005: 349).
One of the clearest cases of pragmatic restrictions is fashion and the “extralinguistic
developments in the society that make certain words or morphological
elements desirable to use” (Plag 1999: 39).
Another pragmatic constraint is the general need for a certain word
(Plag 1999: 39). The need may either relate to the naming or labelling of a new
concept or entity, or to the speakers’ need to condense information due to stylistic
purposes or text cohesion. Yet another, a rather closely related restriction is the
nameability requirement: all new lexemes must denote something nameable in the
language (Plag 1999: 40).
Semantic constraints are not very common, and are mainly
concerned with cases where the referent of the base must follow certain semantic
requirements. For example, the referent of the nouns with the suffix -ee must be
sentient (Barker 1998; cited in Rainer 2005: 349). Semantic restrictions are a
common type of constraint, semantic blocking being one example. Carstairs-
McCarthy defines semantic blocking as a situation in which “the existence of a
word (...) with a particular meaning inhibits the morphological derivation, even by
formally regular means, of another word with precisely that meaning” (2002: 91).
His example makes use of the adjectives curious and glorious. A formally regular
way to form nouns from these adjectives would be to employ the suffix -ity. With
curiosity, this seems to work, the result being curiosity. However, with glorious,
the corresponding noun is glory, not *gloriosity: the existence of gloriosity is thus
blocked by the word glory. Even a formally regular word-formation process can
be blocked (ibid.). One reason for the existence of the phenomenon of semantic
blocking is the tendency to avoid exact synonymy in most languages (Carstairs-
McCarthy 2002: 85).
Rainer calls this kind of blocking token blocking, and writes that
children and non-native speakers often produce formations like gloriosity, which
indicates that the “existence” of a synonymous word refers to the mental lexicon
of the speaker, not to language as a social institution (2005: 336–337). Besides
token blocking, Rainer mentions type blocking, in which case the unacceptability
of a complex word is due to a synonymous pattern that takes precedence (2005:
337–338). He illustrates type blocking by the set of rival suffixes in German,
namely -heit, -ität, and -ie (2005: 338). The suffix -heit is used with adjectives
38

with final stress, and it is fully productive. However, its domain is diminished by -
ität (with learned bases) and, more rarely, -ie (with adjectives ending in -phil).
Férnandez-Domínguez defines both type blocking and token blocking as
examples of synonymy blocking, as opposed to homonymy blocking, which
prevents the coining of a new word because of structural overlapping with an
existing word (2007: 71–73). For example, the word liver ‘inner organ’ prevents
the coining of *liver ‘someone who lives’ (Plag 1999: 50; cited in Férnandez-
Domínguez 2007: 73).
To conclude, it can be said that new words are always formed for a
specific need, and thus constraints may not tell us everything about the
profitability of word-formation processes (Bauer 2001: 143). Férnandez-
Domínguez goes as far as to state that the naming need on the part of the language
community is so strong that it may even “knock out” the restrictions caused by
various constraints (2007: 84). In any case, the application of different wordformation
processes is seldom straightforward, and there are several exceptions to
the rules.
3. Material and methods
3.1. A closer look at the selected combining forms
As was stated earlier, combining forms can be divided into initial and final
combining forms. In this study, I will compare the two and try to find out whether
they differ in productivity. To do so, I will compare the values of Baayen’s
potential productivity P of three initial combining forms (hyper-, ultra-, and
pseudo) as well as three final combining forms (-logy, -graphy, and -nomy), in
order to determine whether there are any major differences in the productivity
rates between these two subtypes. These six combining forms were chosen
because they are among the most frequent ones in the BNC. Sample size is
important in performing statistical analyses on the combining forms, since many
of the measures used in this study do not work properly if the sample is too small
(see section 2.3.3. for discussion).
39

It must be noted that some neoclassical elements have two kinds of
combining forms (with allomorphic variation) derived from the same lexical
word, those that can occur in both initial and final position (cf. Fradin 2000: 13).
For example the Greek λόγο-ς ’word, speech’ occurs in words like logopedics and
biology. A similar example is the Greek form -γράϕος, which is used in
formations like graphology and biography.
Unless otherwise indicated, the OED constitutes the source in the
following description of the selected combining forms. The problem with the
OED, however, is that even though some of the selected combining forms, the
final ones in particular, have rather sketchy entries. McCauley writes that the
amount of information in each entry is to some extent determined by the
frequency, longevity, and complexity of the word or element in the English
language (2006: 99). In addition, most entries are rather old and date back to the
late 19 th century (see section 3.2. for discussion). However, the OED is currently
undergoing a thorough revision that will also affect entries for combining forms:
for example, in the second edition of the OED, there is sometimes ambiguity in
whether a certain element is an affix or a combining form (McCauley 2006: 97).
hyper-
Hyper- is often classified as a prefix instead of a combining form, probably
because it has its origin in the Greek preposition (ὑπέρ ‘over, beyond, over much,
above measure’) instead of a lexical word. Nevalainen classifies hyper- as an
intensifying prefix, and writes that it is the Greek cognate of Latin super-, having
the meaning of ‘over, too much’ (1999: 388). The OED states that hyper- has been
extensively used in English since the 17 th century in the formation of new
compounds (often on Greek analogies), and that it combines relatively freely with
nouns and adjectives.
pseudo-
40

Pseudo- has its origins in the Greek root ψευδο- ‘false, falsity, falsehood’, and is
also quite frequently attached to Latin bases as well. The first attestations of
formations with pseudo- in English date back to early 15 th century in a religious
context, but it did not become common until the 17 th century. The first native
formations (e.g., pseudo-prophetical) date back to that period as well. From the
19 th century onwards, pseudo- combined with nouns and adjectives becomes
frequent in scientific terminology in particular.
ultra-
Unlike the majority of the combining forms in this study, ultra- has its origins in
Latin instead of Greek. It is derived from a Latin preposition that has a meaning
‘beyond’, and like hyper-, it is often classified as a measurement prefix as well.
The original meaning in English is ‘lying spatially beyond or on the other side of’
(as in ultra-terrestrial, ultra-zodiacal), or ‘going beyond, transcending the limits
of’ (ultra-human, ultra-rational). However, from the early 19 th century onwards, a
third sense, ‘an excessive or extreme quality of an adjective’ (ultra-cautious,
ultra-cosmopolitan) has steadily become more common (Hughes 2000: 346–347).
-logy
The origin of -logy is the Greek element λόγο-ς ’word, speech’ or the
corresponding root λόγ-. 19 Derivatives of logy have two distinct senses in English.
The former group includes words with the sense ‘saying or speaking’ (eulogy,
necrology), while the latter sense can be seen in the names of sciences or
disciplines (biology, astrology). All the modern formations of -logy can be said to
have a corresponding formations using the combining forms -logical and -logist
(called suffixal expansion by Prćić (2007: 386–387). Kastovsky states that -logy is
actually much more suffix-like than lexeme-like (2009: 5).
19 Sometimes the thematic vowel -o- is included by dictionary-makers, which results in the form -
ology in English The status of the linking element -o- is discussed in section 2.2.2.
41

-graphy
The form -graphy is closely related to the variant -graph. However, formations
with -graph are not included in this study. The combining form -graphy is derived
from the Greek –γράϕος, which denotes writing. Often the words that have the
component -graphy refer to sciences (geography, bibliography), but they can also
denote writing, drawing, or graphic representation (calligraphy, lithography,
photography). Hybrid formations with -graphy are rare, stratigraphy being one of
the few examples. As with -logy, words with -graphy can have the corresponding
derivatives -graphic and -graphical.
-nomy
The combining form -nomy is derived from the ancient Greek element –νομία, a
variant of the base νέμειν ‘to deal, distribute, hold, manage’. The earliest
attestations of -nomy are astronomy and economy (both from the 16 th century).
There is also a small number of words (from the 16 th century onwards) that denote
‘law’, ‘rule’, or ‘government’, as in autonomy.
3.2. Corpora versus dictionaries in the study of productivity
It was stated above that the most widely-used methods to measure productivity are
either corpus-based or dictionary-based. Both types of methods have their
advantages and disadvantages. Corpora are generally considered to be a more
representative sample of language than dictionaries, but the issue is not so
straightforward, as will be shown.
Historical dictionaries in particular are often partial or even
incomplete, even though they are useful in providing information about the early
attestations of lexemes (Palmer 2009: 8). Nevalainen mentions the various
problems with the OED as regards chronological study of words. For example,
issues such as chronological coverage are often ignored (1999: 337). Therefore,
the citations in the OED do not necessarily reflect the full variety of texts in a
42

given period. For the earliest stages of language, the sampling is often sparse, and
a word may have been in use long before its first written record (Baayen 2008:
909–910). In a similar vein, Schäfer writes that the OED reflects the frequency
and status of a word at the moment of compilation of the dictionary instead of the
period of origin (1989: 69).
According to Baayen and Lieber, some of the benefits of corpora as
opposed to dictionaries are that they reflect actual language usage (1991: 803).
Indeed, as Baayen and Renouf point out, “it is commercially unattractive to print
thousands of words the meaning of which is immediately clear to anyone familiar
with the basic meaning of productive affixes” (1996: 69). According to Booij,
dictionaries only register which new complex words have been established in the
lexicon, while true morphological productivity is reflected by “complex words
that never make it into dictionary”, i.e., hapaxes (2005: 69). In a similar vein,
Baayen and Renouf state that dictionary-based methods may underestimate the
productivity rates of affixes (1996: 70). In addition, the inclusion of a word into a
dictionary is always arbitrary to some extent, which means that a dictionary
cannot be considered as a representative sample of a language (Plag 1999: 27).
The size of the corpus is an important factor in all kinds of linguistic
analysis, and in the study of productivity in particular: in a small corpus, most
hapax legomena will be well-known words of the language, but if the corpus size
increases, the proportion of neologisms among the hapaxes increases as well (Plag
2006: 542–543; see also Baayen and Lieber 1991: 813). For unproductive
processes, increasing the corpus size does not add new types after a certain
“saturation point” has been reached (Lüdeling et al. 2000: 58). In this vein, the use
of the BNC is justified, since it is one of the largest English corpora available.
Other corpora generally considered large enough include the COBUILD / Bank of
English corpus and the COCA corpus of contemporary American English (see
Plag 1999: 25–26). Historical corpora, on the other hand, are often too small for
studying potential productivity P (Palmer 2009: 8–9).
The data for this study was obtained from the written part of the The
British National Corpus (BNC). In total, the BNC comprises just under 100
million word tokens of which roughly 89 million words represent written British
English that falls into various genres, or registers, such as fiction, academic prose,
and personal letters (see Burnard 2007). One of the possible disadvantages with
43

the BNC is the fact that since it does not cover the last twenty years or so, the
most recent changes or trends in the English word-formation are not covered.
However, in this study, the use of the BNC is justified, because it has already been
used in many studies on the productivity of various affixes. This enables the
comparison of the productivity rates to the ones calculated for the selected
combining forms. When necessary, and especially to provide supplementary
information on the behaviour of these combining forms, the Corpus of
Contemporary American English (COCA) will also be consulted, especially in the
discussion section.
It must be noted, however, that a corpus-based productivity measure
is not devoid of problems, either. Lüdeling et al. criticize Baayen for assuming
that all electronic corpora are automatically perfectly prepared (2000: 59). There
are, however, at least three corpus pre-processing problems that must be taken
into consideration: mistagged items, typographical errors, and repetition of the
same texts in corpus composition. Similar problems are even more common in
historical data, since historical corpora often contain several spelling variants of
the same affixes (see Säily and Suomela 2009: 93). Tagging is not a problem in
the present study, since we are only interested in the internal make-up of lexemes.
Even though the BNC has been claimed to have some inherent structural
problems, such as wrongly classified texts and overly broad categories (see Lee
2001), the sampling has been done carefully (see Burnard 2007: section 1.4.4 for
the selection procedures employed), so that the repetition of texts is highly
unlikely.
Typographical errors are, however, a problem that must be taken
seriously. Furthermore, errors often occur only once and consequently add to the
number of hapaxes, and may therefore considerably affect the results (Lüdeling et
al. 2000: 59). Plag expresses the same kind of concern, adding that some
qualitative decisions are necessary even in purely quantitative analysis (1999: 29).
Thus, in order to properly analyse the data for the present study, a certain amount
of manual sorting was necessary.
In general, corpora constitute a better source for the purposes of the
present study. However, as has been shown, corpora as a source are far from
unproblematic. Even the largest, most balanced and representative corpus will
never equal real language use, even though it may provide a useful tool to make
44

certain estimations or predictions based on statistics. As Baroni and Evert have
put it, “corpora are finite samples from the infinite sets that constitutes a
language” (2008: 777). These samples let us make generalizations and inferences,
but this requires that the problem at hand is accurately operationalized in
quantitative terms (Evert and Baroni 2008: 777–778). A good strategy in the study
of productivity is to combine corpus and dictionary evidence to provide a
comprehensive view of the productivity of different patterns (see Plag 2006).
3.3. How to determine units of analysis and other methodological problems
One has to be careful when making inferences based on corpus evidence.
Lüdeling et al. claim that even seemingly error-free corpora may cause trouble in
this respect: for example, words may accidentally contain strings of characters
that look like affixes but are not (2000: 59). For example, not all words that begin
with sub- actually contain the prefix sub- in contemporary English (e.g., subtle),
even if they might be derived from a word that was morphologically complex in
the donor language (Latin sub ‘under’ + tēla ‘web’ ) (Plag 2006: 543). In
psycholinguistics, this phenomenon is referred to as pseudo-affixation (see, e.g.,
Baayen 1993: 199). Another good example of pseudo-affixation is the status of
the constituent-fer in such words as infer, confer, prefer and transfer – it might be
analysed as a bound root, but it does not carry a meaning that would be the same
in all these words (see Plag 2003: 24–25). In the context of the present study,
examples of words that seem to consist of several elements but that actually are
mono-morphemic include hyperion and hyperoid. Therefore, they are excluded
from the type count.
Another similar issue is the fact that many affixes have adopted
idiosyncrasies that may blur their morphological make-up: for example, words
that end in -ity occur in many transparent, complex words, but also in words like
entity, which might not be considered complex (see also Plag 1999: 28). In a
generous count, this word would probably be counted as a unit of analysis and
included in the type count, but in that case it would skew the productivity rate of a
modern word-formation process.
45

Fortunately, the combining forms studied in this paper are quite
distinctive in form, at least compared to affixes in the more traditional sense.
However, several qualitative decisions have to be made concerning which
elements to include and which to exclude from the type count. For example, Prćić
points out that -logy and -holic can be analysed as combining forms in such words
as morphology and foodaholic, but not in apology and alcoholic (2007: 382). In a
similar vein, Warren suggests that hyper- can be considered a prefix in
hyperactive but a combining form in hypertrophy (1990: 113). 20
However, since it is not likely that speakers are aware of such a
distinction, formations such as hypertrophy are not excluded from the data.
Baayen points out that if an affix has several different, although related, senses,
the goodness-of-fit of the word frequency distribution might be lacking (2001:
145). How the decision not to exclude the “pure neoclassical compound”
instances of the selected combining forms affects their productivity values and the
word frequency distributions will be assessed in the Discussion section.
One of the problems of Baayen’s potential productivity index P is
that none of the individual studies that apply the measure explains how the units
of analysis are determined and how the data is treated in the study in question –
this makes it difficult to compare the productivity rates between different authors,
even though they used the same data (see Plag 1999: 29). This is even more
problematic with combining forms (at least if they are to be considered as a wordformation
process distinct from affixation), since methods associated with
affixation might not be fully applied to combining forms.
In his study on the productivity of N + N compounds, Fernández-
Domínguez deals with problems in applying the affix-centred productivity
measures to compounds in English (2009: 142). The problem with compounds is
to determine which part corresponds to the base and which to the affix. The
problem can be tackled by for example calculating the average frequency of the
components (Férnandez-Domínguez 2009: 143). However, in this study I will rely
on the affix-based approach, since native compounds differ from neoclassical
compounds and since combining forms are used in a more affix-like way (i.e.,
they can be combined with free roots as well).
20 In fact,Quirk et al. use the term combining form only with neoclassical compounds, and reserve
the term neoclassical affix for the affix-like use of the same elements (1985: 1545).
46

Determining the appropriate units of analysis is an important issue
as well. For example, whether one counts ultra-violet and ultraviolet as a single
type or separate types has consequences in terms of the P values one obtains.
Sometimes an affix can also have several distinct, although to some extent
interrelated meanings (see, e.g., McConchie 1998 for the different senses of the
prefix dis- in English). Sometimes words generated with an apparently single affix
are in fact generated by several distinct word-formation processes (Lüdeling and
Evert 2003). For example, the German -lich may attach to adjectives, verbs and
nouns alike. Plag also points out that lexicalization, i.e., a process in which a
complex word develops a new, idiosyncratic meaning, may cause problems of
classification (2006: 543). However, the combining forms selected for this study
turned out to be fairly unproblematic in this respect.
In this study, I decided to count the spelling variants of a single
lexeme (quasi-judicial vs. quasijudicial) as tokens of a single type. The spelling
of English compounds is by and large unregulated, so that some compounds can
have as many as three different spelling variants that Férnandez-Domínguez calls
open (pseudo scientific), solid (pseudoscientific), and hyphenated (pseudoscientific)
(2009: 32–33). In a similar vein, words that have different spellings
according to the different national (mainly British vs. American) spelling
standards (e.g., palaeontology vs. paleontology) are counted as a single type.
Of course, the number of types is not as relevant as the number of
tokens in this particular measure of productivity. However, the type hypertension,
for example, occurs only once in the BNC, while hypertension has 369
occurrences. Thus hyper-tension increases the productivity of hyper-, unless it be
counted as a variant of hypertension. For a more detailed discussion of counting
tokens and types, see Baroni (2008: 804–805).
Multiply affixed words constitute another problem in determining
the units of analysis. Should such word as conventionalizable be understood as an
example of -ize or -able – or perhaps both (Plag 1999: 28–29)? The decision one
makes has consequences for the productivity values and word frequency
distributions of the elements selected. In the context of the present study, the
problem is more relevant for final than for initial combining forms.
In the present study, I decided to count formations with a prefix or
an additional initial combining form attached as separate types. Therefore, neuro-
47

physiology (with its spelling variants) and physiology are separate types, as well
as aero-technology and technology. Another way would have been to add up the
frequencies of aero-technology and technology. However, Baayen states that such
cumulated token frequencies are more appropriate in estimating the activation
level A than in calculating the values for P or P* for at least two reasons. First, it
is difficult to estimate the weight of the various word-formation processes in
multiply complex words. Second, if a word is assumed to be an independent free
form in the lexicon, it will be counterproductive to sum up the token frequencies
of these two (1993: 200–201). If a word were assigned to two morphological
categories, the observations in those two categories would no longer be
independent, and the statistical tests for comparing productivity rates would no
longer be valid (Baayen 2008: 903). In their study on the productivity of selected
nominal suffixes (such as -al and -ist), Plag et al. have chosen to exclude prefixed
formations (unavailable) and compounds (performance-artist) completely (1999:
214).
It must be noted again that in some respect the rules concerning
affixes cannot be fully applied to the study of combining forms. For example, it is
not always easy to determine which component is the base and which one
corresponds to the affix. However, it is intuitively clear that adding a prefix (aero-
) to a neoclassical compound (technology) involves lexical creativity, and thus
contributes to the productivity of -logy. On the other hand, since formations with
neoclassical compounds are unclear cases in many ways, the opposite view could
be justified as well.
It must also be noted that there are several final combining forms
that have related forms, such as -logical from -logy and -graphical or -graphic
from -graphy (called suffixal expansion by Prćić (2007: 386–387)). These forms
will not be dealt with in this study, for the reasons mentioned above. P is a
category-conditioned degree of productivity and is thus calculated on the basis of
a single morphological category (see Baayen 1993: 200).
48

4. Results
4.1. Potential productivity P
In order to find all the instances of the initial combining forms, the wildcard * was
used to include all the formations that begin with a certain sequence of letters, in
this case hyper*, ultra*, pseudo*, and quasi*. With respect to final combining
forms, the query was formed as *log[y,ies], *graph[y,ies], etc., in order to include
the plural form of the selected combining forms as well, which were all nounforming.
The queries were restricted to cover only the written part of the BNC.
This makes it easier to compare the results to those obtained in previous studies
(e.g., Plag 1999, 2002, 2006).
Table 2 summarizes the type frequency V, the token frequency N, as
well as the number of hapax legomena n 1 and the values of potential productivity
P of each combining form that was selected for this study.
Table 2. Type frequency V, token frequency N, the number of hapax legomena n 1,
and the productivity rate P of selected combining forms in the written part of the
BNC.
V N n 1 P
hyper- 323 2,859 163 0.0570
ultra- 317 1,628 188 0.1155
pseudo-
325
818
231
0.2824
-logy
375
36,077
111
0.0031
-graphy
166
7,440
57
0.0077
-nomy 46 14,240 20 0.0014
In terms of type frequency V, the initial combining forms generally
seem to rank higher than the final combining forms. The only exception is -logy,
which has the highest type frequency of all the selected combining forms. On the
49

other hand, the final combining form -nomy has a particularly low type frequency,
when compared to the others. It must be noted that the difference in type
frequencies between initial and final combining forms would be even more drastic
if multiply-prefixed words were counted as a single type. In this study, such
words as aero-technology and gastro-technology (both hapaxes) have been
counted as separate types, not as occurrences of technology, and thus the type
frequency (and consequently the number of hapax legomena) is higher than with a
less generous count.
Some tentative conclusions can be already made by looking at the
type frequencies of the selected combining forms, since they clearly measure
some aspect of productivity. As has been stated earlier, this aspect has been called
profitability, extent of use, or realized productivity. Thus, at this point, one could
suggest that -logy has a high extent of use, while with -nomy the extent is not so
substantial. The three initial combining forms and -graphy are situated somewhere
in between on the scale.
As regards the token frequency N, there is more variation, and again
some interesting tendencies can be observed there as well, even though token
frequency in itself has not been generally considered as a very good direct
measure of productivity (see, e.g., Bauer 2001: 147). The most frequent combining
form is -logy, with 36,077 occurrences, which is more than twice as frequent as the
next frequent combining form, which is -nomy with 14,240 occurrences. Both
have some extremely frequent types that contribute greatly to their total token
frequencies. For example, technology has 11,353 occurrences, which equals more
than 30 percent of the total number of tokens. If formations with technology plus a
prefixal element or another initial combining form attached (such as aerotechnology
or multi-technology, both hapaxes) are counted as a single type, the
total token frequency is even higher. As regards -nomy, the relative weight of the
most frequent type (economy with 9,997 occurrences) is even more substantial
(approximately 70 percent of all the tokens with -nomy). In general, the initial
combining forms have far lower token frequencies than the final combining
forms.
In terms of the number of hapax legomena n 1, the initial combining
forms display much higher frequencies than the final combining forms. The
hapaxes also include many lexemes that are not attested in the OED. These
50

include such items as hyperalertness, pseudo-Scottish, and ultra-Zionist. As for
the final combining forms, such hapaxes include ghettology and dreamography.
There are at least two reasons why they are not listed in the dictionary. Firstly,
they are quite rare in the language, and secondly, they are semantically rather
transparent. For example, hyperalertness need not be listed in a dictionary,
because alertness already has an entry, and because its meaning can be constituted
from the meaning of its parts. This can be seen as another indicator for the
productivity of the combining forms examined, since productivity and
transparency are at least indirectly linked (see Dalton-Puffer 1994: 252–253,
Pounder 2000: 134). As was stated earlier, transparent, infrequent words are
parsed, instead of being accessed as a whole (Frauenfelder and Schreuder 1992).
The only exception in this group of words is -nomy, since all hapaxes with -nomy
have an entry in the OED. Besides, most of them are scientific words, such as
chirognomy and aeronomy.
At this point, it might be justified to examine the ratio of the number
of hapaxes to the overall type frequency of each combining form. This ratio has
not been used in any of the previous literature as a measure of morphological
productivity, but it might offer some tentative insights on the relative importance
of hapaxes. In this sense, it certainly describes some aspect of productivity by
being indirectly linked to potential productivity P.
Table 3 displays the proportion of hapax legomena out of the total
type frequency for each combining form. The ratio is higher with initial
combining forms, being over 50 per cent in all the cases. The final combining
forms do not seem to lag very far behind, however. The table can help make
predictions about the P values of the combining forms, which will be discussed
next.
51

Table 3. The ratio of hapax legomena n 1 to the type frequency V of selected
combining forms in the written part of the BNC.
n 1 / V
hyper- 0.5046
ultra- 0.5931
pseudo-
-logy
-graphy
0.7108
0.296
0.3434
-nomy 0.4347
The last column in Table 2 presents the P values for the selected
combining forms. Again, the initial combining forms score higher. The lower P
values of the final combining forms are partly explained by the presence of those
few very frequent types mentioned above. With the initial combining forms, such
extremely frequent types are nonexistent. For example, with hyper-, which is the
least productive initial combining form as regards potential productivity P, the
most frequent type is hypertension (368 occurrences) and with ultra-, ultraviolet
(332 occurrences). These are only a fraction of the corresponding numbers for the
final combining forms.
What can be said about the overall productivity of the selected
combining forms? The problem with interpreting the results is that the P values
alone cannot give us any definitive answers. Baayen and Lieber emphasize the
fact that P is only a relative measure and cannot therefore in itself be used to
determine whether a particular word-formation rule is productive or not (1991:
816). The method is statistical, but it does not give any threshold values that
would indicate the boundaries between productive and unproductive processes. As
Plag has noted, P only provides “a continuum of more or less productive
processes”, so that it is difficult or even impossible to pin down where exactly the
productive processes end and the unproductive processes begin (1999: 33).
Baayen and Lieber suggest that in order to decide whether a process
is productive, its P value should be compared to the P value of a relevant set of
52

simplex words (1991: 816). However, calculating the P value of, say, an
inflectional process would be beyond the scope of this study. Another way to
estimate the productivity of the selected combining forms is to compare it to a set
of words formed with another word-formation process, such as suffixes as
opposed to prefixes, or vice versa. Plag, for example, provides the P values for
seven different suffixes in English (2006: 545, based on data from Plag et al. 1999
and Plag 2002). The data come from the written part of the BNC as well (see Plag
et al. 1999: 210), so the figures should be more or less comparable to those of the
present study, even though Plag does not offer very rigorous details about the
basis on which he has determined the units of analysis (see section 3).
Table 4. Type frequency V, token frequency N, the number of hapax legomena n 1,
and the productivity rate P of selected suffixes in the written part of the BNC.
From Plag (2006: 545).
V N n 1 P
-ness 2,466 1,369,116 943 0.061
-ion 2,392 371,747 524 0.0338
-ity 1,372 106,957 354 0.0096
-ist
1,207
98,823
341
0.0088
-less
682
28,340
272
0.0036
-ish
491
7,745
262
0.00092
-wise 183 2,091 128 0.00038
Table 4 presents the type and token frequencies as well as the
number of hapax legomena and the potential productivity values of selected
suffixes in the written part of the BNC, calculated by Plag (2006: 545). The suffix
-wise has the highest P value (0.061), and -ion the lowest (0.00038). If these
values are compared to those of the combining forms that were examined in this
study, it can be seen that the suffixes have significantly lower P values than the
combining forms do. Even the least productive prefix, hyper-, (P = 0.0570) has a
53

somewhat higher P value than the most productive suffix in Plag’s study. As for
the final combining forms, they score lower than the most productive suffixes, but
higher than the least productive ones. One of the disadvantages with Plag’s study
is that is since it does not account for prefixes, it is not possible to see whether
prefixes in general would rank higher than suffixes.
To summarize, the initial combining forms seem to rank somewhat
higher than the final combining forms, both in terms of the number of different
types V and potential productivity P. On the other hand, the final combining forms
in general have higher token frequencies N, which is often caused by the existence
of a few very frequent types. What this section has discussed constitutes the basic
productivity measures that have been established and widely adopted in the study
of productivity. There is, however, a set of measures more rarely employed in
studies on the productivity of different word-formation processes. These measures
rely on lexical statistics and word frequency distributions, and thus illustrate a
slightly different aspect of productivity.
4.2. LNRE modelling
It was stated above that Zipf(-Mandelbrot)’s law fits the word frequency
distribution of naturally occurring texts reasonably well. The same holds true for
productive word-formation processes as well, which makes LNRE modelling
particularly useful for the study of productivity. There are several methods that
can be used to obtain information on the productivity of the combining forms
selected.
As was stated above, one of the inherent problems with P is that the
proportion of hapax legomena does not grow linearily as the corpus size N
increases. This makes it difficult to compare the P values between different wordformation
processes, since the sample sizes for each process varies. A solution
that was offered in the previous section was to compare the sets of P values of
different word-formation processes. However, there is a more elegant way to
determine the productivity of the combining forms: the zipfR package of the R
statistics software (Evert and Baroni 2007) allows the prediction of the shape of
54

the vocabulary growth curve at arbitrary sample sizes. 21 Another method that is
particularly useful in the study of productivity is calculating the expected
frequency spectra for the combining forms and comparing them to the observed
values. This allows us to assess the goodness-of-fit of the statistical model, i.e.,
how well it fits the data. Therefore, it can help us detect possible outliers, i.e., data
points that deviate significantly from the rest of the sample.
Table 5 presents the observed and expected P values for each
combining form. For -logy, the values are identical, since -logy has the highest
token frequency and therefore serves as the “base” against which the other
combining forms are compared. For the other combining forms, the differences
between the observed and expected values vary, depending on their sample size.
In addition to their P values, the ordering of the combining forms after the
extrapolation operation changes as well. However, pseudo- has the highest P
value in both cases, and -nomy the lowest.
Table 5. The observed and expected P values for the selected combining forms in
the written part of the BNC.
P
P (expected)
(observed)
hyper- 0.0570 0.0052
ultra- 0.1155 0.0047
pseudo-
-logy
-graphy
0.2824
0.0031
0.0077
0.0347
0.0031
0.0028
-nomy 0.0014 0.0007
21 For a sample session illustrating the necessary operations for obtaining the expected P values for
different word-formation processes, see Evert and Baroni 2007.
55

A more visual way to compare the productivity rates is to look at
their vocabulary growth curves, which are shown in figure 4. A vocabulary
growth curve can be used to show changes in vocabulary size V as the sample size
N increases (Evert and Baroni 2007: 32). The steeper the curve, the more rapidly
the vocabulary (i.e., the number of types) is growing, and the more productive a
particular word-formation process can be considered. In a sense, a vocabulary
growth curve works in a similar way to the potential productivity P, since both
attempt to capture the rate at which the vocabulary of a given word-formation
process grows. The only difference is that a vocabulary growth curve relies on
types, while P makes use of hapax legomena.
As can be seen in Figure 4, pseudo- holds the highest place, which is
shown by the steepest curve. On the other hand, -nomy has the lowest
productivity, which is represented by a moderate, almost flat curve. In general, the
initial combining forms retain the highest productivity ranks, while all the final
combining forms cluster at the low end of the graph. All the slopes follow the
general tendency to grow more rapidly at low N and then gradually decrease. At
small sample sizes, it is more likely to encounter a yet unattested type, but the
likelihood decreases as the sample size increases.
56

Figure 4. Extrapolated vocabulary growth curves for the selected combining
forms. The x axis displays corpus size N, and the y axis the expected type
frequencies at different sample sizes.
57

Figure 5. The expected (black) versus observed (red) frequency spectra for the selected
combining forms. The x axis displays frequency classes, and the y axis displays the
number of types in each class.
58

Not only vocabulary growth curves, but also frequency spectra
provide a useful tool to assess the productivity of different word-formation
processes. In particular, a frequency spectrum provides information on the
distribution of the type frequencies of a given word-formation process, because
productive ones tend to behave in a certain way in this respect, producing a
skewed, rapidly decreasing plot. Even looking at the frequency spectra per se can
reveal interesting facts, but it is even more useful to compare the observed values
to the expected, average values. Again, this can be done with the ZipfR package
of the R statistics software. The actual formula to calculate the expected values for
a frequency spectrum is explained by Baayen (2001: 92).
Figure 5 shows the expected (black) versus observed (red) frequency
spectra for the combining form data. The figure only displays the first fifteen
frequency classes, but it can nevertheless give a useful visualization of how the
different types are distributed with each combining form. Even a brief glance at
the figure shows that all the combining forms display similar, rapidly decreasing
plots with a long tail of high frequency classes that are realized by only few types.
In addition, as the figure shows, almost all the observed frequency spectra line up
relatively beautifully with the expected, average values.
The only exception is -nomy, which displays quite drastic
differences between the expected and the observed values at times. For example,
frequency class 2 (i.e., the number of types that occur twice in the corpus) has
many more members than expected. The same holds true for frequency class 5.
Scarcity of data cannot be the reason for this inconsistency, since the token
frequency of -nomy is the second largest of all the selected combining forms. On
the other hand, despite its high token frequency, -nomy has a particularly low type
frequency, only 46 types (the most frequent combining forms having over 300
types). In a situation like this, even a small deviation from the expected values
might have an impact. Another striking feature with -nomy, which might have an
impact on the results is that the most frequent type, economy, accounts for over 70
per cent of all the words with -nomy. With -graphy, the corresponding percentage
is roughly 20% (the most frequent type being geography), and with hyper, no
more than 13% (hypertension).
59

One explanation for the lack of goodness-of-fit might also be
sampling error, in which case the irregularities would be outliers, i.e., atypical
data points with exceptionally high or low values (see Baayen 2008: 91).
However, one has to be careful when discarding data as potential outliers. As
Baayen has put it, “you should not tweak the data by removing data points so that
a non-significant effect becomes significant” (2008: 237). Baayen also points out
that the statistical LNRE models rely on the assumption that words occur
randomly and independently of each other in texts, which is of course a
simplification and might cause the lack of goodness-of-fit (2008: 233). Of course,
the randomness factor might not be as pervasive in lexical studies as in studies on
syntax.
The slight out-of-datedness of the BNC mentioned in section 3.2 is a
factor that must not be overlooked either. In fact, there are several cases where a
more up-to-date corpus would certainly have yielded different results. The word
euro-cracy offers a good example, even though it is not among the combining
forms selected for this study. It is a hapax in the BNC, but in view of the extralinguistic
development of the last two decades, it is certainly not a very rare word
in British English. On the other hand, eurocracy also turned out to be a hapax in
the COCA corpus, which is updated every year and should thus cover even the
most recent changes in the English language. However, this might partially be
explained by the fact that the COCA records American English. Therefore the outof-date
nature of the BNC shows up and skews the results. It is also possible to
find examples of words which have been fashionable at the moment of corpus
compilation, but which have since then become rare or even obsolete, as a result
of changes in fashion or technological development. Hypersparc (100
occurrences), which refers to a 1990’s microprocessor, is a good example.
5. Discussion
It is a relatively easy task to extract figures and tables from the data and count the
P values for a set of word-formation processes – even if the methodological
problems discussed in section 3.3 are taken into consideration. In a similar vein,
60

calculating frequency spectra and vocabulary growth curves with R is a relatively
straightforward process. In fact, in many studies that rely on P as an indicator of
productivity, the discussion of results is rather superficial and the evidence drawn
from the P values of affixes is taken for granted.
However, careful qualitative analysis of the results is necessary in
order to get a realistic view of the productivity of these combining forms.
According to Evert and Lüdeling, quantitative and qualitative analyses can
complement each other, and while quantitative approaches provide useful
information on the nature of different word-formation processes, they cannot
really be used without qualitative interpretation (2000: 167–168). There are
several linguistic and extra-linguistic factors that must be taken into account. The
factors may also shed light on eventual differences in productivity rates obtained
with various productivity measures. It is also worthwhile to discuss the reasons
that might explain those differences. Issues like these will be the topic of this
section.
In general, all the measures employed in this study seem to indicate
that the initial combining forms have higher productivity rates than the final
combining forms. They have higher P values, and they produce steeper
vocabulary growth curves. On the other hand, it seems that there are no major
differences in the goodness-of-fit of the frequency spectra between the initial and
the final combining forms comparing the observed and expected frequency
spectra. However, based on the evidence brought about by potential productivity
P and vocabulary growth curves, we might well claim that that the initial
combining forms score higher than the final combining forms.
What might the reasons for the higher productivity rates obtained by
the initial combining forms be? The differences between initial and final
combining forms from a semantic or structural point of view have not been
considered in previous literature, but the prefix-suffix distinction might prove a
useful analogy, since it is not certain that speakers make a distinction between
affixes and combining forms. Nevalainen writes that prefixes have fewer wordclass
restrictions on their input range than suffixes do (1999: 355). From the
semantic point of view, the main linguistic function of prefixes is often semantic,
while suffixes tend to be class-changing, but more abstract in meaning (see
section 2.1.).
61

Studies concentrating on or touching upon the productivity of
combining forms have mainly been theoretical. Warren, for example, rather
cryptically writes that prefixes have “productive force”, while combining forms
“need not have productive force” but can be nonce formations (1990: 123). 22
Similarly, Prćić states that the productivity of prefixes is systematic, since they are
used regularly in “ready-made morpho-syntactic patterns”, and that each prefix
can be assigned a value of high, restricted, or low productivity (2005: 325).
Combining forms, on the other hand, “are simply there to be used if/when need
for them arises”, and thus display non-systematic productivity, comparable to that
of elements in compounding (Prćić 2005: 325). Fischer states that dictionaries can
be used to assess the productivity of combining forms since if a combining form
has its own entry in a dictionary, it can be said to have at least some degree of
productivity (1998: 63). However, the fact that all the elements studied in this
paper have an entry in the OED is too weak an indicator in itself to gauge their
productivity.
It is also worthwhile to consider the general tendencies or
underlying changes that might be going on in terms of the English word-formation
system. Bauer argues that since many of the native prefixes in English have lost
ground, having been replaced by learned “prefix-like elements”, a gradual
typological shift from prefixation towards “something more like compounding”
might be taking place (2003a: 35, 37). Even though he does not state it explicitly,
by these “learned prefix-like elements” Bauer seems to specifically mean
neoclassical combining forms. Szymanek seems to hold a similar view as well,
claiming that the neoclassical compounds have gained ground in the English
word-formation system, especially during the last few decades (2005: 432).
However, Bauer and Szymanek do not discuss whether these tendencies apply to
suffixes and final combining forms as well.
Szymanek also points out that coining new words, at least nouns, by
compounding, is easy for speakers, due to recursion and the absence of any major
grammatical restrictions (2005: 432). The same holds true for prefixes, as they,
22 Bauer (1983: 45–46) defines nonce formations as complex words that are created “on the spur of
the moment”, usually to cover an immediate need. Thus they are usually unique and rare. Nonce
formations must not be confused with hapax legomena: hapaxes are always defined in terms of a
certain corpus. When nonce formations start to be used regularly in the speech community and
become institutionalized, they become neologisms (Fischer 1998: 15–16).
62

too, can be used recursively, and usually do not affect the stress pattern of their
bases and do not have segmental phonological effects on their bases (Lieber 2005:
389).
In a similar vein, as mentioned earlier, Hughes has noted that many
learned prefixes, such as extra-, super- and hyper- have gained new, creative
meanings and are thus used with greater flexibility in contemporary English than
they were before (2003: 346–347). Yet another fact suggesting the high
productivity of prefixes or initial combining forms comes from Bauer, who uses
the term word families to illustrate the fact that a single prefixing operation may
result in several members of the same family (as in biodegradability,
biodegradable, biodegrade, biodegradation), which of course increases the P
value of the element in question. Kastovsky notes that besides scientific
terminology, neologisms that exploit combining forms also occur in science
fiction novels, which also demonstrates the productivity of this word-formation
process (2009: 1). He mentions such examples as cryosleep, holomovie, and exobiology.
With these observations in mind, combined with evidence from the BNC,
it is not too bold to state that all the combining forms studied in this paper are
fairly productive.
Etymology is a factor that might play an important role in the
differences between the productivity rates between initial and final combining
forms. The latter all come from lexical words (mostly nouns), while some initial
combining forms come from prefixes (hyper) – do they resemble prefixes in other
senses as well? Another question is how readily the selected combining forms
combine with native bases as opposed to Greek and Latin elements – a possible
topic for a future study.
All this raises the question whether speakers of English actually
differentiate between affixes and combining forms in general and whether the
distinction between these two word-formation processes has any relevance from
the point of view of cognitive processing and the mental lexicon. As has already
been demonstrated above, it is not always easy to distinguish between affixes and
combining forms (or combining forms and bases) even in terms of morphological
theory. The distinction between prefixes and initial combining forms in particular
does not seem to be so clear-cut. Prćić states that modern morphology in general
has been unable to make a consistent distinction between these two types of word-
63

formation, and they are also often labelled inconsistently in dictionaries as well
(2005: 313–314).
Marchand writes that learned affixes and combining forms are “on
the same footing” in the English word-formation system, since they are both
borrowed elements that do not exist independently as words (1969: 131–132).
Similarly, Bauer, when suggesting that English is in the process of going through
a gradual typological change, in fact defines prefixes and initial combining forms
as a single category of “prefixal elements” and treats elements like anti-, sub-,
super-, mini- and micro- as members of a similar category, although the OED
defines the first three as prefixes, and the latter two as combining forms (2003a).
In a similar manner, Nevalainen seems to treat all the prefixal elements, native
and non-native alike, as a single category that she calls prefixes (1999: 383).
Prćić suggests that certain elements that have traditionally been
defined as combining forms (including pseudo- and quasi-) may become more
like affixes after repeated use: they acquire a “ready-made morphosemantic
pattern” (2005: 329). 23 He goes as far as to suggest that elements like mega-,
micro-, multi-, pseudo- and quasi-should be recategorized as prefixes for various
reasons, one of them being the fact that they modify the meaning of the base in a
regular way (2005: 329). Warren has come to a similar conclusion, claiming that
affixes traditionally result from a grammaticalization process while combining
forms were originally bound lexical elements und thus “better equipped to resist
grammaticalization” (1990: 123–124). However, she also writes that combining
forms are not immune to grammaticalization either, and thus some of them have
moved towards an affixal status. Fundamentally, this all comes down to
Kastovsky’s claim (2009) that that a category like combining forms is altogether
unnecessary and that processes like affixation, compounding, blending and
clipping could be used to deal with that category.
In any case, it is evident that productivity in the strict sense only
measures one aspect of productivity. A more comprehensive study would involve
a combination of different methods. Baayen and Lieber themselves acknowledge
the fact that the type frequency of a given affix is also an indicator of its extent of
23 Carstairs-McCarthy writes that the recent proliferation of combining forms in association with
free Germanic roots can actually be seen as a rather peculiar trend, since Latin- and Greek-derived
words are otherwise quite rare in present-day English (2002: 109).
64

use and may thus be of interest in such fields as lexicography or applied
linguistics (1991: 817).
6. Conclusion
Sometimes ignorant but pretentious people take to
coining words, re-interpreting foreign words in their own
way. They vaguely feel that there is some characteristic
termination in a Greek or Latin word which they then
attach to some English basis to give a combination a
‘learned’ tinge. As a result, we get barbarisms in -athon,
coined after Marathon, such as danceathon, swimathon,
etc... (Marchand 1969: 213, emphases orig.)
Whether Marchand’s rather extreme claim is justified or not, he may well
consider this battle as lost. What was considered “a purely dictionary interest”
(see Marchand 1969: 132) a few decades ago, is now an integral part of the
English lexicon without doubt. The main results of this study indicate that what
has here been called combining forms are indeed frequent in English, and they
combine quite freely with both native and non-native elements.
This study has attempted to measure the productivity of six
neoclassical combining forms (three initial and three final) in English, using
several complementary methods. First, a quantitative, statistical measure P
developed by Baayen and his co-workers has been applied to these combining
forms. Secondly, two measures of lexical statistics, i.e., vocabulary growth curves
and frequency spectra, have been used in order to take as many aspects of
productivity into account as possible. The motivation for using these methods has
been discussed, as well as the various problems that have occurred during the
research process and the appropriate solutions.
As regards their P values, the initial combining forms seemed to
score somewhat higher than the final combining forms. In terms of vocabulary
growth curves, the initial combining forms produced steeper curves than the final
combining forms, which indicates that their vocabulary grows at a greater pace
65

and that they are therefore more productive. The third measure, i.e., the
comparison of observed and expected frequency spectra, did not yield any major
differences between the initial and the final combining forms. The observed and
the expected values lined up rather nicely with almost all of the selected elements.
The findings are not surprising, given the latest tendencies in
English word-formation that seem to favour prefix-like elements over suffixes. It
has even been suggested that a typological shift might be under way in the English
word-formation system (Bauer 2003a). In addition, the proliferation of scientific
texts seems to contribute to the high productivity of these elements. In order to
make broader generalizations, though, a larger set of combining forms should be
compared. However, the problem is that in performing an analysis that draws
from lexical statistics, sample size matters: many combining forms do not produce
many hits even in a large corpus like the BNC. This makes their statistical analysis
hard. R might fail to produce expected frequency spectra for some combining
forms if the sample size is too small. For example, the initial combining form
ultra- has 2000 occurrences in the BNC, which is not enough for R to produce an
extrapolated vocabulary growth curve.
It must also be noted that most of the combining forms selected
turned out to be rather affix-like by nature, their function being quantifying rather
than that of resembling a lexeme. This is probably related to their high frequency
in English, as opposed to more lexeme-like combining forms. However, these
combining forms with more clearly lexeme-like meanings (such as bio- ‘life’ or
astro- ‘star’) would probably have acquired different, probably lower, P values.
Another important aspect that must be kept in mind at this point is
psycholinguistics. Whether speakers actually differentiate between affixes and
combining forms from the point of view of mental lexicon and cognitive
processing is a question worth investigating. As Plag has noted, “speakers of
English can and do master English morphology without etymological knowledge”
(2002: 286). It is the boundary between prefixes and initial combining forms that
seems to be rather blurry. Final combining forms, on the other hand, seem to be
more distinct from affixes as a category.
During the research process, it has also become evident that
Baayen’s quantitative method measures only one aspect of productivity, i.e., the
growth rate of the vocabulary of a certain morphological category (see Baayen
66

2008: 902). As Plag has put it, it is “only a more or less accurate statement of the
problem and not a solution to it” (1999: 35). Therefore the potential productivity
P cannot be the only way to assess the productivity of the selected prefixes and
combining forms. In this study, methods of lexical statistics were used to offer an
account that is as diverse as possible and that takes into account the different
aspects of morphological productivity.
Another interesting topic for future research could involve
comparing the productivity rates of combining forms across different registers,
following the example of Plag et al. (1999). Many Greek prefixes in particular are
relatively recent loans in English and are mainly used in word-formation in
specialized scientific registers (Biber et al. 1999: 324). The problem with that kind
of research is that the samples compared must be of equal size, which, as Säily
has noted, may lead to discarding valuable data (2011: 125). On the other hand,
studying the correlation of productivity and different sociolinguistic variables,
such as gender, age or social rank of the speakers (see Säily and Suomela 2009,
Säily 2011), is also worth further study. Another aspect of productivity that has
not been thoroughly investigated yet is the competition between different wordformation
processes that express the same need to name various concepts and
phenomena (see Lüdeling et al. 2006).
In the future, more theoretical accounts of productivity are also
needed: lately, studies of productivity have mainly concentrated on the
productivity of individual word-formation processes (see Dressler and Ladányi
2000: 104). Besides, even though the development of different empirical and
statistical methods to measure productivity has flourished in the past few years,
there are still many questions that concern the very nature of productivity that
need to be tackled. Plag, for example, mentions such problems as allomorphy and
the relationship between rival morphological processes, as well as how
morphology interacts with phonology, syntax, and semantics (1999: 1). In
addition, the study of productivity might have practical applications beyond
morphology, measuring the “generative potential” (Evert/Baroni) of such things
as syntactic processes or collocations/idioms. LNRE models in particular are also
an area that has not so far been widely employed in the study of productivity.
Kastovsky argues that instead of considering the demarcation
between combining forms and affixation, a more relevant question would actually
67

e the difference between compounding and affixation in general (2009: 10–11).
He points out that a number of native affixes has developed from members of
compounds (e.g., -dom, -hood, and -wise), and the same phenomenon can be seen
with combining forms (2009: 10). Kastovsky suggests that the theoretical
framework of grammaticalization would probably be productive in this kind of
analysis (2009: 11). From the semantic point of view, a crucial distinction in this
model is also the one between “lexically specific” and “lexically non-specific”
word-formation processes, which can also be seen as a cline between words and
stems on the one hand and affixes on the other hand (Kastovsky 2009: 11).
Another research topic not yet fully covered is the so-called modern combining
forms, such as jazzo-(phile) or speedo-(meter) (Prćić 2005, 2007, 2008).
Primary sources and software
BNC = The British National Corpus, version 4.1 (BNC XML Edition). 2008.
Distributed by Oxford University Computing Services on behalf of the
BNC Consortium. Online: http://www.natcorp.ox.ac.uk/
COCA = The corpus of contemporary American English. Compiled by M. Davies.
2008– Online: http://corpus.byu.edu/coca
The R project for statistical computing. http://www.r-project.org
References
Amiot, D. and G. Dal. 2007. Integrating neoclassical combining forms into a
lexeme-based morphology. In G. Booij.et al. (eds.), On-line proceedings
of the fifth Mediterranean Morphology Meeting (MMM5) Fréjus 15–18
September 2005, University of Bologna. Online:
http://mmm.lingue.unibo.it/mmm-proc/MMM5/232-336-Amiot-Dal.pdf
[accessed on October 26, 2012]
68

Anshen, F. and M. Aronoff. 1981. Morphological productivity and phonological
transparency. The Canadian Journal of Linguistics 26 (1), 63–72.
Aronoff, M. 1976. Word formation in generative grammar. Cambridge, MA: MIT
Press.
Aronoff, M. and F. Anshen. 1998. Morphology and the lexicon: Lexicalization
and productivity. In A. Spencer and A.M. Zwicky (eds.), The Handbook
of morphology. (Blackwell Handbooks in Linguistics). Cambridge, MA:
Blackwell Publishers, 237–248.
Baayen, R. H. 1992. Quantitative aspects of morphological productivity. In G.
Booij and J. van Marle (eds.), Yearbook of morphology 1991.
Dordrecht: Kluwer, 109–149.
Baayen, R. H. 1993. On frequency, transparency and productivity. In G. Booij and
J. van Marle (eds.), Yearbook of morphology 1992. Dordrecht: Kluwer,
182–208.
Baayen, R.H. 2001. Word frequency distributions. Dordrecht: Kluwer.
Baayen, R.H. 2003. Probabilistic approaches to morphology. In R. Bod et al.
(eds.), Probabilistic linguistics. Cambridge, MA, MIT Press, 229–287.
Baayen, R. H. 2008. Analysing linguistic data: a practical introduction to
statistics using R. Cambridge: Cambridge University Press.
Baayen, R. H. 2009. Corpus linguistics in morphology: morphological
productivity. In A. Lüdeling and M. Kytö (eds.), Corpus linguistics: An
international handbook, ed. by Anke Lüdeling and Merja Kytö. Berlin:
Walter de Gruyter, 899–919.
Baayen, R. H. and R. Lieber. 1991. Productivity and English derivation: A
corpus-based study. Linguistics 29 (5), 801–843.
Baayen, R. H. and A. Renouf. 1996. Chronicling the Times: Lexical innovations
in an English newspaper. Language 72 (1), 69–96.
Barker, C. 1998. Episodic -ee in English: a thematic role constraint on new word
formation. Language 74: 695–727.
Baroni, M. 2008. Distributions in text. In A. Lüdeling and M. Kytö (eds.), Corpus
linguistics: an international handbook. Berlin: Mouton de Gruyter,
803–822.
Baroni, M. and S. Evert. 2008. Statistical methods for corpus exploitation. In A.
Lüdeling and M. Kytö (eds.), Corpus linguistics: an international
handbook. Berlin: Mouton de Gruyter, 777–803.
69

Bauer, L. 1983. English word-formation. (Cambridge Textbooks in Linguistics).
Cambridge: Cambridge University Press.
Bauer, L. 1988. Introducing linguistic morphology. Edinburgh: Edinburgh
University Press.
Bauer, L. 1992. Scalar productivity and -lily adverbs. In G. Booij and J. van Marle
(eds.), Yearbook of morphology 1991. Dordrecht: Kluwer, 185–191.
Bauer. L. 1995. Is morphological productivity non-linguistic? Acta Linguistica
Hungarica 43 (1–2), 19–31.
Bauer, L. 1998. Is there a class of neoclassical compounds and if so is it
productive? Linguistics 36 (3), 403–422.
Bauer, L.. 2001. Morphological productivity. (Cambridge Studies in Linguistics).
Cambridge: Cambridge University Press.
Bauer, L. 2003a. English prefixation – a typological shift? Acta Linguistica
Hungarica 50 (1–2), 33–40.
Bauer, L. 2003b. Introducing linguistic morphology. 2nd edition. Edinburgh:
Edinburgh University Press.
Bauer, L. 2003c.The productivity of (non-)productive morphology. Rivista di
Linguistica 15, 7–16.
Bauer, L. 2005 Productivity: theories. In Štekauer and R. Lieber, 315–334.
Biber, D. et al. 1999. Longman grammar of spoken and written English. London:
Longman.
Bloch, B. and G.L. Trager. 1942. Outline of linguistic analysis. Baltimore:
Linguistic Society of America.
Booij, G. E. 2005. The grammar of words: an introduction to linguistic
morphology. Oxford: Oxford University Press.
Burnard, L. (ed.). 2007. Reference guide for the British National Corpus (XML
edition). Published for the British National Corpus Consortium by the
Research Technologies Service at Oxford University Computing
Services. Online: http://www.natcorp.ox.ac.uk/docs/URG/ (accessed on
1 October 2012).
Burnley, L. 1992. Lexis and semantics. In Blake, N. (ed.) The Cambridge history
of the English language, Vol. 2: 1066–1476. 409–499.
Carstairs-McCarthy, A. 1992. Current morphology. London/New York:
Routledge.
70

Carstairs-McCarthy, A. 2002. An introduction to English morphology: words and
their structure. (Edinburgh Textbooks on the English Language).
Edinburgh: Edinburgh University Press.
CEEC = Corpus of Early English Correspondence. 1998. Compiled by Terttu
Nevalainen, Helena Raumolin-Brunberg, Jukka Keränen, Minna
Nevala, Arja Nurmi and Minna Palander-Collin at the Department of
English, University of Helsinki.
Chitashvili, R.J. and H. Baayen 1993. Word frequency distributions. In L.
Hrebícek and G. Altmann (eds.) Quantitative text analysis. Trier :
WVT, 54-135.
Corbin, D. 1987. Morphologie dérivationnelle et structuration du lexique.
Tübingen: Max Niemeyer.
Cottez, H. 1985. Dictionnaire des structures du vocabulaire savant: éléments et
modèles de formation. Paris: Le Robert.
Cowie, C. 1998. The discourse motivations for neologising: Action
nominatization in the history of English. In J. Coleman and C. J. Kay
(eds.), Lexicology, semantics and lexicography. Selected papers from
the fourth G. L. Brook Symposium.(Amsterdam Studies in the Theory
and History of Linguistic Science. Series 4, Current issues in linguistic
theory, vol. 194). Amsterdam: John Benjamins.179–207.
Cowie, C. 2003. “Uncommon terminations”: Proscription and morphological
productivity. Rivista di Linguistica 15 (1), 17–30.
Cowie, C. and C. Dalton-Puffer. 2002. Diachronic word-formation and studying
changes in productivity over time: Theoretical and methodological
considerations. In J. E. Díaz Vera (ed.) A changing world of words:
Studies in English historical lexicography, lexicology and semantics.
Amsterdam: Rodopi.410–437.
Dalton-Puffer, C. 1994.Productive or not productive? The Romance element in
Middle English derivation. in F. Férnandez et al. (eds.) English
Historical Linguistics 1992: Papers from the 7th International
Conference on English Historical Linguistics, Valencia, 22–26
September 1992. Amsterdam: John Benjamins.247–260.
Dalton-Puffer, C. 1996.The French influence on Middle English morphology: A
corpus-based study of derivation. (Topics in English Linguistics 20).
Berlin: Mouton de Gruyter.
71

Dressler, W. U. and M. Ladányi. 2000. Productivity in word formation (WF): A
morphological approach. Acta Linguistica Hungarica 47 (1–4), 103–
144.
Evert, S. and A. Lüdeling. 2001. Measuring morphological productivity: Is
automatic preprocessing sufficient? In P. Rayson et al. (eds.)
Proceedings of the Corpus Linguistics 2001 Conference. Lancaster:
University Center for Computer Research on Language. 167–
175.Online:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.28.9162&rep
=rep1&type=pdf [accessed on October 15, 2012]
Evert, S. and A. Lüdeling. 2003. Linguistic experience and productivity: Corpus
evidence for fine-grained distinctions. In D. Archer et al. (eds.)
Proceedings of the Corpus Linguistics 2003 Conference. UCREL
technical paper 16. Lancaster: UCREL. 475-483.Online:
http://citeseer.ist.psu.edu/viewdoc/similar?doi=10.1.1.13.2813&type=ab
[accessed on January 13, 2013]
Evert, S. and M. Baroni. 2007. zipfR: Word frequency distributions in R.
Proceedings of the 45 th Annual Meeting of the Association of
Computational Linguistics, Posters and Demonstrations Session.
Prague, Czech Republic. Online: http://www.stefanevert.de/PUB/EvertBaroni2007.pdf
[retrieved on April 13, 2013]
Férnandez-Domínguez, J. et al. 2007. How is low morphological productivity
measured? Atlantis 29 (1), 29–54.
Férnandez-Domínguez, J. 2009.Productivity in English word-formation: An
approach to n+n compounding. (European University Studies, series
21, Linguistics, vol. 341) New York: Peter Lang.
Fischer, R. 1998. Lexical change in present-day English: A corpus-based study of
the motivation, institutionalization, and productivity of creative
neologisms. (Language in Performance). Tübingen: G. Narr.
Fradin, B. 2000. Combining forms, blends and related phenomena. In U.
Doleschal and A. M. Thornton (eds.), Extragrammatical and
marginal morphology. Munich: LINCOM Europa, 11–60.
Frauenfelder, U. H. and R. Schreuder 1992.Constraining psycholinguistic models
of morphological processing and representation: the role of
productivity. In G. E. Booij and J. van Marle (eds.), Yearbook of
morphology 1991. Dordrecht: Kluwer Academic Publishers, 165–183.
Harastani, R. et al. 2012. Neoclassical compound alignments from comparable
corpora. In A. Gelbukh (ed.), Computational linguistics and intelligent
72

text processing 13 th international conference, CICLing 2012, New
Delhi, India, March 11–17, 2012, Proceedings, Part II. Heidelberg:
Springer, 72–82.
Haspelmath, M. 2002. Understanding morphology.(Understanding Language
Series). London: Arnold.
Hay, J. 2001. Lexical frequency in morphology: Is everything relative?
Linguistics 39, 1041–1070.
Hay, J. 2003. Causes and consequences of word structure. (Outstanding
Dissertations in Linguistics). New York: Routledge.
Hay, J. and H. Baayen. 2002. Parsing and productivity. In G. Booij and J. van
Marle (eds.), Yearbook of morphology 2002, 203-235. Dordrecht:
Kluwer.
Hohenhaus, P. 2005. Where word-formation cannot extend the vocabulary:
Creativity, productivity and the lexicon in synchronic and diachronic
morphology. In D. Kastovsky and A. Mettinger (eds.), Lexical change
and the genesis of English vocabulary. Berlin/New York: Mouton de
Gruyter,
Huddleston, R. and G. Pullum. 2002. The Cambridge grammar of the English
language. Cambridge: Cambridge University Press.
Hughes, G. 2003. A history of English words. Oxford: Blackwell Publishing.
Kastovsky, D. 2001. English morphology: A typological reappraisal. In C.
Schaner-Wolles et al. (eds.). Naturally! Linguistic studies in honour of
Wolfgang Ulrich Dressler presented on the occasion of his 60th
birthday. Turin: Rosenberg and Sellier.215–224.
Kastovsky, D. 2006. Typological changes in derivational morphology. In A. van
Kemenade and B. Los (eds.), The handbook of the history of English.
(Blackwell Handbooks in Linguistics). Malden: Blackwell Publishing.
151–176
Kastovsky, D. 2009. Astronaut, astrology, astrophysics: About combining forms,
classical compounds and affixoids. In R. W. McConchie et al. (eds.)
Selected proceedings of the 2008 Symposium on New Approaches in
English Historical Lexis (HEL-LEX 2). Somerville, MA: Cascadilla
Proceedings project. 1-13. Online: http://www.lingref.com/cpp/hellex/2008/paper2161.pdf
[accessed on January 14, 2013].
Krott, A. et al. 1999. Complex words in complex worlds. Linguistics 37. 905–926.
73

Lee, D. 2001. Genres, registers, text types, domains, and styles: clarifying the
concepts and navigating a path through the BNC jungle. Language
Learning & Technology 5 (3), 37–72.
Lehrer, A. 1998.Scapes, holics, and thons: The semantics of English combining
forms. American Speech 73 (1), 3–28.
Lieber, Rochelle. 2005. English word-formation processes: Observations, issues,
and thoughts on future research. In Štekauer and Lieber (eds.), 375–427.
Lipka, L. 1990. An outline of English lexicology. (Forschung & Studium Anglistik
3). Tübingen: Niemeyer.
Lüdeling, A. et al. 2000. On measuring morphological productivity. Proceedings
of the KONVENS 2000, 57–61. Online:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.106.4819&re
p=rep1&type=pdf [accessed on November 16th, 2012].
Lüdeling, A. et al. 2002. Neoclassical word-formation in German. In Booij.G. and
J. van Marle (eds.) Yearbook of morphology 2001.Dordrecht: Kluwer,
253–283.
Lüdeling, A. and S. Evert. 2003. Linguistic experience and productivity: corpus
evidence for fine-grained distinctions. Proceedings of the Corpus
Linguistics 2003 Conference. Online: http://www.ims.unistuttgart.de/projekte/corplex/paper/evert/LuedelingEvert2003.pdf
[accessed on November 26, 2012].
Lüdeling, A. et al. 2006. Need and competition: Deconstructing quantitative
productivity. Online: http://cogsci.uniosnabrueck.de/~qitl/abstracts/luedeling_baroni_evert.pdf
[accessed on
November 25, 2012].
Marchand, H. 1969. The categories and types of present-day English wordformation:
A synchronic-diachronic approach. 2nd edition. Munich:
Beck’sche.
Marle, J. van. 1985. On the paradigmatic dimension of morphological creativity.
(Publications in language sciences 15). Dordrecht: Foris Publications.
McCauley, J. 2006. Technical combining forms in the third edition of the OED:
Word-formation in a historical dictionary. In R. W. McConchie et al.
(eds.) Selected proceedings of the 2005 Symposium on New Approaches
in English Historical Lexis (HEL-LEX).Somerville, MA: Cascadilla
Proceedings Project. 95–104. Online: http://www.lingref.com/cpp/hellex/2005/paper1350.pdf
[accessed on January 14, 2013]
74

McConchie, R. W. 1998. The vernacularization of the negative prefix dis- in Early
Modern English. Lexicology, semantics and lexicography. In J.
Coleman and C. J. Kay (eds.) Selected papers from the fourth G. L.
Brook Symposium.(Amsterdam Studies in the Theory and History of
Linguistic Science. Series 4, Current issues in linguistic theory, vol.
194) Amsterdam: John Benjamins. 209–227.
McConchie, R. W. 2006. Disseisin: The lexeme and the legal fact in Early Middle
English. In R. Dury el al. (eds.) English historical linguistics 2006:
Lexical and semantic change. Amsterdam: John Benjamins. 203–216.
Minkova, D. and R. Stockwell. 2009. English words: Structure and history. 2nd
edition. Cambridge: Cambridge University Press.
Miller, D.G. 2012. External influences on English: From its beginnings to the
renaissance. Oxford: Oxford University Press.
Nevalainen, T. 1999. Early Modern English lexis and semantics. In Lass, R. (ed.).
The Cambridge history of the English language, Vol. 3: 1476–1776.
Cambridge: Cambridge University Press. 332–458.
OED = The Oxford English dictionary online.2007 (Online version 2013).3rd
edition. Oxford: Oxford University Press.Online: http://www.oed.com
Palmer, C. C. 2008. Borrowed derivational morphology in Late Middle English: a
study of the records of the London Grocers and Goldsmiths. In
Fitzmaurice, S. et al. (eds.). Studies in the history of the English
language IV: empirical and analytical advances in the study of English
language change. (Topics in English linguistics 61). Berlin/New York:
Mouton de Gruyter, 231-64.
Palmer, C. C. 2009. Borrowings, derivational morphology, and perceived
productivity in English, 1300–1600.Ph. D. thesis. University of
Michigan. Online:
http://deepblue.lib.umich.edu/bitstream/2027.42/64624/1/palmercc_1.pd
f [retrieved September 18, 2012]
Pápai, V. 2004. Explicitation: A universal of translated text? In A. Mauranen and
P. Kujamäki (eds.) Translation universals: Do they exist? Amsterdam:
John Benjamins Publishing.143-164.
Plag, I. 1999. Morphological productivity. Structural constraints in English
derivation.(Topics in English Linguistics 28). Berlin: Mouton de
Gruyter.
75

Plag, I. 2002. The role of selectional restrictions, phonotactics, and parsing in
constraining suffix ordering in English. In G. Booij and J. van Marle
(eds.), Yearbook of morphology 2001, 285–314.Dordrecht: Kluwer.
Plag, I. 2003, Word-formation in English. (Cambridge Textbooks in Linguistics).
Cambridge: Cambridge University Press.
Plag, I. 2006. Productivity. In B. Aarts and A. McMahon (eds.), The handbook of
English linguistics. (Blackwell Handbooks in Linguistics). Malden:
Blackwell Publishing.537–556.
Plag, I. et al. 1999. Morphological productivity across speech and writing. English
Language and Linguistics 3 (2), 209–228.
Pounder, A. 2000. Processes and paradigms in word-formation morphology.
Berlin: Mouton de Gruyter.
Prćić, T. 2005. Prefixes vs. initial combining forms in English: A lexicographic
perspective. International Journal of Lexicography 18 (3), 313–334.
Prćić, T. 2007. Headhood of suffixes and final combining forms in English wordformation.
Acta linguistic hungarica 54 (4), 381–392.
Prćić, T. 2008. Suffixes vs. final combining forms in English: A lexicographic
perspective. International Journal of Lexicography 21 (1), 1–22.
Quirk, R. et al. 1985. A comprehensive grammar of the English language.
London: Longman.
Rainer, F. 2005. Constraints on productivity. In Štekauer and Lieber (eds.), 335-
352.
Renouf, Antoinette. 2007. Tracing lexical productivity and creativity in the British
Media: ‘The Chavs and the Chav-nots’. In J. Munat (ed.) Lexical
creativity. texts and contexts. (Studies in Functional and Structural
Linguistics). Amsterdam: Benjamins. 61–89.
Säily, T. 2011. Variation in morphological productivity in the BNC:
Sociolinguistic and methodological considerations. Corpus Linguistics
and Linguistic Theory 7 (1), 119–141.
Säily, T. and J. Suomela. 2009. Comparing type counts: The case of women, men
and -ity in early English letters. In A. Renouf and A. Kehoe (eds.)
Corpus linguistics: Refinements and reassessments. (Language and
Computers: Studies in Practical Linguistics 69.) Amsterdam: Rodopi.
871–909. Authors’ version available from
http://www.cs.helsinki.fi/u/josuomel/doc/icame-2007-paper.pdf
[retrieved on October 27, 2012]
76

Schäfer, J. 1989. Early Modern English: OED, New OED, EMED. In R. W.
Bailey (ed.), Dictionaries of English: prospects for the record of our
language. Cambridge: Cambridge University Press. 66-74.
Štekauer, P. and R. Lieber (eds.) 2005.Handbook of word-formation. Dordrecht:
Springer.
Štekauer, P. et al. 2005.Word-formation as creativity within productivity
constraints: sociolinguistic evidence. Onomasiology Online 6, 1–55.
Online:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.92.9286&rep
=rep1&type=pdf [Accessed on November 18, 2012]
Szymanek, B. 2005. The latest trends in English word-formation. In Štekauer and
Lieber (eds.), 429–448.
Tweedie, F. J. and R. H. Baayen. 1998. How variable may a constant be?
Measures of lexical richness in perspective. Computers and the
Humanities 32, 323–352.
Warren, B. 1990.The importance of combining forms. In W. U. Dressler et al
(eds.) Contemporary morphology.Trends in linguistics: Studies and
monographs 49. Berlin: Mouton de Gruyter.111–132.
Zipf, G. K. 1949. Human behaviour and the principle of least effort. Cambridge,
MA: Addison-Wesley.
Zipf, G. K. 1965. The psycho-biology of language. Cambridge, MA: MIT Press.
77